Analysis method and device for electricity market transaction agent decision behavior

Abstract

The invention provides an analysis method and device for a power market transaction agent decision behavior. The analysis method comprises the following steps: in an agent decision process of a power transaction simulation platform, obtaining a price-electric quantity pair generation number of each segment, a proportion of a consumed step number of a current inference chain and an actual coverage rate of a high-price interval segment; identifying an inference resource insufficient interval in which the ratio of the remaining available inference step number to the high-price interval coverage rate is lower than a preset threshold value so as to determine a segmentation granularity division scheme needing to be adjusted; and according to a segmentation granularity division scheme needing to be adjusted, re-dividing the generation tasks of each price interval under the upper limit constraint of the step number of proxy single-round reasoning.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method and apparatus for analyzing the decision-making behavior of agents in the electricity market. Background Technology

[0002] In the field of electricity market trading, studying agent decision-making behavior is crucial for optimizing resource allocation and improving market efficiency. Electricity trading is not only a core link in energy allocation but also an important mechanism for ensuring stable power supply and economic benefits. However, how to formulate reasonable bidding strategies in a complex trading environment directly affects the returns of market participants and the overall system performance. Currently, many solutions often overlook the practical operational limitations of agents with limited computing resources when designing bidding strategies. Although increasing the number of segments in the bidding curve can more closely reflect real cost changes, this approach does not fully consider the boundaries of computing power, leading to new problems in strategy execution. Especially under resource constraints, too many segments may prevent the decision-making process from fully covering all key price ranges, thus affecting the comprehensiveness and accuracy of the bidding. A deeper technical challenge lies in the irreconcilable contradiction between the number of segments in the bidding curve and the allocation of computing resources. Increasing the number of segments significantly increases the need for fine-grained characterization of price changes, but it also significantly increases the number of price and power combinations that need to be processed in each round of decision-making. This makes it difficult for agents to complete a full analysis of all price ranges within limited computational steps. High-price ranges, in particular, are often overlooked or hastily processed due to resource exhaustion, resulting in incomplete or biased bids. For example, in actual electricity trading, agents need to generate bids for multiple price ranges within a short period based on market fluctuations. If there are too many segments, the agent may only complete bids for low-price ranges before exhausting computational resources, failing to provide reasonable solutions for high-price ranges. This leads to an inability to effectively compete when market electricity prices are high, resulting in missed profit opportunities or disruption of market balance. Summary of the Invention

[0003] The purpose of this invention is to provide a method and apparatus for analyzing the decision-making behavior of agents in the electricity market, aiming to solve the problem that the existing technology cannot provide a reasonable pricing scheme in actual electricity transactions, resulting in too many segments that may prevent the decision-making process from fully covering all key price ranges.

[0004] On the one hand, this invention provides a method for analyzing the decision-making behavior of agents in electricity market transactions, mainly including: In the agent decision-making process of the power trading simulation platform, obtain the number of price-electricity pairs generated in each segment, the percentage of steps consumed in the current inference chain, and the actual coverage of the high-price range. The remaining available inference steps are analyzed based on the percentage of steps consumed in the current inference chain. When the ratio of the remaining available inference steps to the actual coverage of the high-price range is lower than a preset ratio threshold, the inference resource shortage range where the number of generated price-power pairs in the high-price range is lower than the generation number threshold is obtained. Based on the inference resource shortage range, the segmentation granularity division scheme that needs to be adjusted is determined. Based on the segmentation granularity scheme that needs to be adjusted, the generation tasks for each price range are redefined under the upper limit constraint of the number of inference steps in a single round of proxy, and the price-electricity pair generation distribution is obtained based on the generation task. Based on the price-electricity pair generation distribution, identify the incomplete quotation intervals in the high-price intervals that have not yet been generated due to exhaustion of steps, reallocate the number of segments to the high-price intervals, and calculate the segment granularity of each interval after adjustment. Based on the adjusted segmented granularity of each interval, the generation order and step allocation of each price interval in the agent inference chain are reconfigured to obtain the updated price curve generation sequence. This sequence is then applied in the agent decision to generate price-electricity pairs. The adjusted generated price sequence is evaluated to determine whether it has covered the upper limit of the price, thus obtaining the complete generated price curve.

[0005] Furthermore, during the agent decision-making process of the power trading simulation platform, the number of price-electricity pairs generated in each segment, the percentage of steps consumed in the current inference chain, and the actual coverage of the high-price range are obtained, including: The agent captures the price-energy pairs output in each price segment in real time, records the number of price-energy pairs generated in each segment, arranges them in order from low to high price axis to form a segment generation number sequence, and marks the price boundary value corresponding to each segment to obtain a segment generation statistics table. Based on the segmented statistical table, the number of inference steps executed in the current inference chain of the agent and the preset single-round inference step limit are read, and the percentage of steps consumed in the current inference chain is calculated. When the percentage of consumed steps exceeds the step percentage threshold and the number of generated high-price intervals is lower than the number of generated intervals threshold, the high-price interval is marked as an under-covered interval. Based on the location identifiers of the under-covered intervals, coverage statistics are performed on the high-price intervals to obtain the actual coverage rate of the high-price intervals.

[0006] Furthermore, the remaining available inference steps are analyzed based on the percentage of steps consumed in the current inference chain. When the ratio of the remaining available inference steps to the actual coverage of the high-price range is lower than a preset ratio threshold, an inference resource shortage range is identified where the number of price-power pairs generated in the high-price range is lower than a generation number threshold. Based on the inference resource shortage range, an adjustment to the segmentation granularity scheme is determined, including: Obtain the remaining available inference steps in the agent's current inference chain and the actual coverage of the high-price range. Divide the remaining available inference steps by the actual coverage to obtain the step coverage ratio. When the step coverage ratio is lower than the preset ratio threshold, the high-price interval is marked as an inference resource shortage interval, and the starting price boundary and ending price boundary of the inference resource shortage interval on the price axis are recorded. The price span of the inference resource shortage interval is determined based on the starting price boundary and the ending price boundary on the price axis. The current segment granularity is calculated based on the price span of the inference resource shortage interval and the number of currently allocated segments. Based on the comparison between the remaining available inference steps and the estimated number of steps required to complete full coverage of the interval, the degree of resource gap is determined, and the segment boundaries of the interval with insufficient inference resources are re-divided according to the degree of resource gap, resulting in an adjusted segmentation granularity scheme.

[0007] Furthermore, the step of re-dividing the generation task for each price range according to the segmentation granularity partitioning scheme that needs adjustment, under the constraint of the upper limit of the number of steps in a single round of agent inference, and obtaining the price-electricity pair generation distribution based on the generation task, includes: Extract the adjusted number of segments for each price range from the segmentation granularity division scheme that needs adjustment, and simultaneously read the maximum number of steps for a single round of agent inference; Arrange the price ranges in ascending order of price to form a price range sorting sequence; For each price range, the adjusted number of segments is used to divide the maximum number of steps in a single round of inference according to the proportion of the number of segments in each range to the total number of segments, thus obtaining the step quota table for each price range. Based on the step quota table for each price range, the task of generating price-energy pairs is executed sequentially according to the segment boundaries within each price range. After the price-energy pairs for each segment are output, the number of generated pairs for that segment is recorded. The distribution of generated price-energy pairs is obtained by summarizing them according to the price axis position.

[0008] Furthermore, the generation order and step allocation of each price interval in the agent inference chain are reconfigured based on the adjusted segmented granularity of each interval to obtain an updated price curve generation sequence. This sequence is then applied in the agent decision-making process to generate price-electricity pairs. The process evaluates whether the adjusted generated price sequence has covered the price ceiling, resulting in a fully generated price curve, including: Obtain the segmentation granularity and segmentation number allocation table for each interval after adjustment, and arrange each price interval in order of price from low to high; For each price range, a corresponding inference step quota is allocated. The inference step quota is calculated based on the number of segments in the range and the average number of inference steps per segment, forming a generation order and step allocation configuration table for each price range. Construct a quotation curve generation sequence based on the generation order and step allocation configuration table; During the agent decision-making process, the agent executes the pricing task for each price range sequentially according to the pricing curve generation sequence. The agent outputs price-electricity pairs in each segment and records the position of the price range that has been output. Determine whether the highest price in the current price sequence has reached the preset price limit. If it has, then the price sequence has completed the coverage of the entire price range, resulting in a complete price curve.

[0009] Furthermore, the price curve generation sequence includes the segment boundaries, segment granularity, and corresponding step quotas for each price range.

[0010] Furthermore, the method also includes: The Monte Carlo simulation algorithm is used to evaluate the fitting accuracy and strategy exploration completeness of the quoted price curve. The results of the simulation are analyzed to improve the coverage of the high-price range. Feedback data on agency behavior in electricity market transactions are obtained to identify the improvement effect of the quoted price distortion in the feedback data. The segmented granularity adjustment parameters of the price range are iteratively updated to determine the inference resource allocation scheme.

[0011] Furthermore, the Monte Carlo simulation algorithm is used to evaluate the fitting accuracy and strategy exploration completeness of the price curve, analyze the coverage improvement results of the high-price range in the simulation results, obtain feedback data on agency behavior in electricity market transactions, identify the improvement effect of price distortion in the feedback data, and iteratively update the segmented granularity adjustment parameters of the price range to determine the inference resource allocation scheme, including: The Monte Carlo simulation algorithm is used to perform multiple rounds of random sampling on the fully generated price curve, simulating price fluctuation scenarios in the electricity market in each round of sampling; The price curve is compared with the market clearing price in the simulated scenario, and the deviation between the price of each segment and the market clearing price is calculated to obtain the fitting accuracy index of the price curve. The proportion of times each price range was covered by sampling out of the total number of samplings was used to obtain the strategy exploration completeness index; Based on the fitting accuracy index and the strategy exploration completeness index, the number of times the high-price interval is covered in multiple rounds of simulation is extracted. The current coverage rate is compared with the coverage rate before adjustment to obtain the coverage improvement result of the high-price interval. Record the agent's pricing output behavior and corresponding market response data in each simulation round to form an agent behavior feedback dataset; From the agent behavior feedback dataset, identify segments where the deviation between the quoted price and the market clearing price exceeds a preset deviation threshold, mark them as price distortion segments, and count the change in the number of price distortion segments before and after adjustment to obtain the improvement effect of price distortion. When the improvement effect does not reach the preset improvement threshold, the number of segments in the corresponding interval is increased according to the distribution position of the price distortion segments, and the step quota ratio of the high-price interval is adjusted according to the coverage improvement result. The segment granularity adjustment parameters are iteratively updated to obtain the inference resource allocation scheme.

[0012] On the other hand, the present invention provides an analysis device for the decision-making behavior of electricity market transaction agents, mainly comprising: The data acquisition module is used to acquire the number of price-electricity pairs generated in each segment, the percentage of steps consumed in the current inference chain, and the actual coverage of the high-price segment during the agent decision-making process of the power trading simulation platform. The resource identification module is used to analyze the remaining available inference steps based on the proportion of steps consumed in the current inference chain. When the ratio of the remaining available inference steps to the actual coverage of the high-price interval is lower than a preset ratio threshold, the module obtains the inference resource shortage interval where the number of generated price-power pairs in the high-price interval is lower than the generation number threshold, and determines the segmentation granularity division scheme that needs to be adjusted based on the inference resource shortage interval. The task partitioning module is used to re-divide the generation tasks of each price range according to the segmentation granularity partitioning scheme that needs to be adjusted, under the constraint of the upper limit of the number of inference steps in a single round of the agent, and obtain the price-electricity pair generation distribution based on the generation task. The interval adjustment module is used to identify incomplete price intervals in the high-price intervals that have not yet been generated due to exhaustion of steps, based on the price-electricity pair generation distribution, reallocate the number of segments to the high-price intervals, and calculate the segment granularity of each interval after adjustment. The sequence configuration module is used to reconfigure the generation order and step allocation of each price interval in the agent inference chain based on the segmented granularity of each interval after adjustment, obtain the updated price curve generation sequence, apply the sequence to generate price-electricity pairs in the agent decision, evaluate whether the adjusted generated price sequence has covered the price upper limit, and obtain the complete generated price curve.

[0013] Furthermore, the device also includes: The evaluation iteration module is used to evaluate the fitting accuracy and strategy exploration completeness of the price curve using the Monte Carlo simulation algorithm, analyze the coverage improvement results of the high-price range in the simulation results, obtain feedback data on agency behavior in electricity market transactions, identify the improvement effect of price distortion in the feedback data, and iteratively update the segmented granularity adjustment parameters of the price range to determine the inference resource allocation scheme.

[0014] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a method and apparatus for analyzing agent decision-making behavior in electricity market transactions. Addressing the problem of incomplete bid curve generation, particularly low coverage of high-price ranges, due to insufficient agent inference resources in electricity trading, a systematic solution is proposed. This invention deploys a bid generation monitoring and inference chain integrity verification module to acquire real-time data on price-electricity pair generation distribution and inference step consumption, identify resource-deficient ranges, and optimize segmentation granularity and step allocation based on dynamic programming algorithms and Monte Carlo simulations. It also reconfigures the inference chain generation order to ensure physically achievable coverage of the entire price range within the step limit. The core innovation of this invention lies in significantly improving high-price range coverage through iterative adjustment of segmentation granularity parameters. Simultaneously, it utilizes simulation results to analyze the improvement effect on bid distortion and obtain agent behavior feedback data, ultimately achieving accurate fitting of the bid curve and complete strategy exploration, providing efficient and reliable decision support for electricity market transactions. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating an analytical method for agent decision-making behavior in electricity market transactions, as described in this invention.

[0016] Figure 2 This is a schematic diagram of the structure of an analysis device for agent decision-making behavior in electricity market transactions according to the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.

[0018] Please refer to Figure 1 , Figure 1 The flowchart illustrates a method for analyzing the decision-making behavior of agents in electricity market transactions according to an embodiment of the present invention, specifically including: Step S101: In the agent decision-making process of the power trading simulation platform, obtain the number of price-electricity pairs generated in each segment, the percentage of steps consumed in the current inference chain, and the actual coverage of the high-price segment.

[0019] In the power trading simulation platform, the price-electricity pairs output by the agent in each price segment are captured in real time. The number of price-electricity pairs generated in each segment is recorded, and the generated numbers are arranged in ascending order of price axis to form a segment generation number sequence. At the same time, the price boundary value corresponding to each segment is marked to obtain a segment generation statistics table. Based on the segment generation statistics table, the percentage of steps consumed in the current inference chain is calculated by reading the number of inference steps executed in the agent's current inference chain and the preset single-round inference step limit (defined as 100 steps). If the percentage of steps consumed exceeds the step percentage threshold and the number of generated high-price intervals (defined as intervals with price boundary values ​​higher than 80 yuan / kWh) is lower than the generation number threshold, then the high-price interval is marked as an under-covered interval, and the location identifier of the under-covered interval is output. Using the location identifier of the under-covered area, the number of segments that have been generated and the number of segments that have not been generated in the high-price range are counted. The number of high-price segments that have been generated is divided by the total number of segments in the high-price range to obtain the actual coverage rate of the high-price range. The sequence of the number of segments generated, the number of price-power pairs generated in each segment, the percentage of steps consumed, and the actual coverage rate are used as the monitoring output of the agent decision-making process.

[0020] In the power trading simulation platform, when the agent executes the quotation generation task, the price-electricity pairs output by the agent in each segment are captured one by one according to the division boundary of the price segment.

[0021] For example, if the price axis is divided into several continuous intervals, after the agent completes the quotation output for a certain interval, the number of price-electricity pairs generated in that interval is immediately recorded, and the generated number is arranged in order of price from low to high to form a segmented generation number sequence.

[0022] In one possible implementation, the process of constructing the segmented generation statistics table includes: for each price segment, marking its corresponding lower and upper price boundaries; associating and storing the price boundary values ​​with the number of segments generated, thereby forming a segmented generation statistics table with price boundary identifiers. The verification task is then performed based on the segmented generation statistics table, with the following working mechanism: reading the number of inference steps already executed in the agent's current inference chain, and simultaneously obtaining the preset upper limit for the number of inference steps per round (system configuration parameter, set to 10); dividing the number of inference steps already executed by the upper limit for the number of inference steps per round to obtain the percentage of steps consumed.

[0023] It should be noted that the step percentage threshold and the number of generated prices threshold are two independent judgment conditions. The former is used to measure the degree of inference resource consumption, while the latter is used to measure the sufficiency of price generation in the high-price range. When the percentage of consumed steps exceeds the step percentage threshold, and the number of price-power pairs generated in the high-price range is lower than the number of generated prices threshold, the high-price range is marked as an under-coverage range, and the position identifier of the under-coverage range on the price axis is output. Based on the position identifier of the under-coverage range, the coverage rate of the high-price range is statistically analyzed. The statistical process includes traversing all segments within the high-price range, distinguishing between segments that have been generated and those that have not yet started generating prices, and dividing the number of high-price segments that have been generated by the total number of segments in the high-price range to obtain the actual coverage rate of the high-price range. The number of price-power pairs generated in each segment, the percentage of consumed steps, and the actual coverage rate together constitute the monitoring output of the agent decision-making process, which is used to reflect the agent's price generation status for the entire price range under the current inference resource constraints.

[0024] Step S102: Analyze the remaining available inference steps based on the percentage of steps consumed in the current inference chain. When the ratio of the remaining available inference steps to the actual coverage of the high-price segment is lower than a preset ratio threshold, obtain the inference resource shortage interval where the number of generated price-power pairs in the high-price segment is lower than the generation number threshold. Determine the segmentation granularity division scheme to be adjusted based on the inference resource shortage interval.

[0025] The remaining available inference steps in the agent's current inference chain and the actual coverage rate of the high-price range are obtained. The remaining available inference steps are divided by the actual coverage rate to obtain the step coverage ratio. If the step coverage ratio is lower than a preset ratio threshold, the high-price range is marked as an inference resource shortage range. The starting price boundary and ending price boundary of the inference resource shortage range on the price axis are recorded. Based on the starting price boundary and ending price boundary of the inference resource shortage range, the price span of the range is calculated. At the same time, the number of currently allocated segments in the range is obtained. The price span is divided by the number of allocated segments to obtain the current segment granularity. The remaining available inference steps are compared with the estimated number of steps required to complete full coverage of the range to obtain the resource gap value. Based on the resource gap value, the degree of resource gap is determined. The degree of resource gap is divided into mild gap, moderate gap, and severe gap according to the ratio of the resource gap value to the estimated number of steps. Using the resource gap degree and the current segmentation granularity, the segmentation boundaries are redefined for the inference resource shortage interval. The number of segments is adjusted according to the correspondence that a mild gap corresponds to reducing one segment, a moderate gap corresponds to reducing two segments, and a severe gap corresponds to reducing three segments, to obtain the adjusted segmentation granularity and form the segmentation granularity division scheme that needs to be adjusted.

[0026] In power trading simulation scenarios, calculating the step coverage ratio is a key step in identifying intervals where inference resources are insufficient.

[0027] Specifically, when an agent executes a pricing task, the remaining available inference steps in its inference chain reflect the amount of computing resources currently available for allocation, while the actual coverage rate of the high-price range reflects the degree of completion of the pricing task within that range. Dividing the remaining available inference steps by the actual coverage rate yields a step-coverage ratio, which represents the remaining steps of resources corresponding to a unit of coverage. When this ratio is lower than a preset threshold, it indicates that the remaining steps are insufficient to support complete coverage of the high-price range, and this high-price range is marked as an inference resource-insufficient range.

[0028] In one embodiment, the boundary records of the inference resource shortage interval include the starting price boundary and the ending price boundary of the interval on the price axis. These two boundary values ​​are used to determine the price range of the interval. The process of determining the degree of resource gap involves comparing the remaining available inference steps with the estimated number of steps required to complete full coverage. The estimated number of steps is calculated based on the number of currently allocated segments within the interval and the average number of inference steps consumed per segment. The average number of steps consumed is derived from historical inference task data and calculated using the formula A=T / N, where A is the average number of steps, T is the total number of historical steps, and N is the number of historical segments. The resource gap value is defined as the difference between the estimated number of steps and the remaining available inference steps, which reflects the number of steps lacking to complete full coverage of the interval. Based on the ratio of the resource gap value to the estimated number of steps, the degree of resource gap is divided into three levels: when the ratio is below 0.3, it is determined to be a mild gap; when the ratio is between 0.3 and 0.6, it is determined to be a moderate gap; and when the ratio exceeds 0.6, it is determined to be a severe gap. This hierarchical mechanism ensures that the adjustment range of segment granularity matches the degree of resource scarcity. Based on the aforementioned resource shortage level, the segment boundaries are redefined for intervals with insufficient inference resources.

[0029] In one possible implementation, a slight gap corresponds to reducing one segment, a moderate gap corresponds to reducing two segments, and a severe gap corresponds to reducing three segments. By reducing the number of segments, the price span of each segment within the inference resource shortage interval increases accordingly, forming an adjusted segment granularity, thereby obtaining the segment granularity division scheme that needs to be adjusted.

[0030] Step S103: Based on the segmentation granularity division scheme that needs to be adjusted, the generation tasks of each price range are re-divided under the upper limit constraint of the agent's single round inference steps, and the price-electricity pair generation distribution is obtained based on the generation task.

[0031] Obtain the segmentation granularity partitioning scheme that needs adjustment, extract the adjusted number of segments for each price range from the scheme, and simultaneously read the upper limit of the agent's single-round inference steps. Use this upper limit as the total number of steps allocated to the generation task for each price range, and arrange the price ranges in ascending order of price to form a price range sorting sequence. Based on the price range sorting sequence, allocate a corresponding inference step quota for the adjusted number of segments within each price range. The allocation method for the inference step quota is to divide the upper limit of the single-round inference steps according to the proportion of the number of segments in each range to the total number of segments, resulting in a step quota table for each price range. This step quota table records the inference step resources obtained for each price range. Using the step quota table, execute the price-electricity pair generation task sequentially according to the segment boundaries within each price range. After each segment's price-electricity pair is output, record the number of generated segments. Summarize the number of generated segments within each price range according to the price axis position to obtain the price-electricity pair generation distribution.

[0032] In the power trading simulation scenario, the segmented granularity partitioning scheme refers to the way power prices are subdivided. Obtaining this scheme is a prerequisite for re-partitioning and generating tasks. This task adjustment aims to optimize simulation efficiency.

[0033] Specifically, the adjusted number of segments for each price range (i.e., ranges divided by price range, such as 0-100 yuan / kWh) is extracted from the segmentation granularity scheme that needs adjustment (derived from the initial simulation settings). Simultaneously, the upper limit of the agent's single-round inference steps is read as the total number of steps constraint. The price ranges are arranged in ascending order of price, forming a price range sorting sequence. This sequence clarifies the execution order of each price range in the generation task.

[0034] It should be noted that the allocation of step quotas follows the principle of proportional division. For each price range in the price range sorting sequence, the corresponding inference step resources are allocated from the single-round inference step limit based on the proportion of its adjusted number of segments to the total number of segments.

[0035] For example, if the number of segments in a certain price range accounts for one-quarter of the total number of segments, then the inference step quota obtained by that range is one-quarter of the maximum number of inference steps in a single round. Through this proportional allocation method, the step quota of each price range is recorded in a step quota table, which contains the identifier of each price range, the number of segments, and the corresponding inference step resources.

[0036] In one possible implementation, a price-electricity pair generation task is performed based on the step quota table. This generation task is executed by an agent model, with the price segment boundaries and electricity demand as inputs, and the corresponding price-electricity pairs as outputs. Within each price range, the task is executed sequentially from the lower price segment to the higher price segment according to the segment boundaries. After each segment's price-electricity pairs are output, the number of generated pairs for that segment is recorded, and each generation consumes a fixed number of steps (1). When the step quota for a certain price range is exhausted, the generation task in that range is stopped, and execution continues in the next price range. Further, the number of segments generated in each price range is summarized according to the price axis position (i.e., the midpoint price of the segment) to obtain the price-electricity pair generation distribution. This distribution forms a histogram with the price axis as the horizontal axis and the number of generated pairs for each segment as the vertical axis, reflecting the agent's price coverage status for each price range under the constraint of the upper limit of single-round inference steps.

[0037] Step S104: Based on the price-electricity pair generation distribution, identify the incomplete quotation intervals in the high-price intervals that have not yet been generated due to exhaustion of steps, reallocate the number of segments to the high-price intervals, and calculate the segment granularity of each interval after adjustment.

[0038] Obtain the price-electricity pair generation distribution, traverse the generation count of each segment in the high-price range, and if the generation count of a segment is zero and the segment is located in a price range after the step quota is exhausted, then mark the segment as a price incomplete interval, and count the total number of price incomplete intervals. Based on the total number of price incomplete intervals and the current generation count of each segment in the low-price range, a dynamic programming algorithm is used to redistribute the number of segments. The dynamic programming algorithm uses the upper limit of the single-round inference steps as a constraint and aims to maximize the number of covered price ranges. From the low-price range, segments with a generation count exceeding the average generation count of the range are selected as redundant segments, and a portion of the number of segments in the redundant segments is transferred to the price incomplete intervals, resulting in a segment number allocation table for each price range. Using the segment number allocation table, the segment granularity is recalculated for each price range. The segment granularity is calculated by dividing the price span of the range by the adjusted number of segments. Simultaneously, the number of segments in each price range is multiplied by the average inference steps per segment, and these are summed to obtain the adjusted total inference chain length. If the adjusted total inference chain length exceeds the step limit, the price range with the smallest segment granularity is selected from the segment number allocation table. The number of segments in this range is reduced, and the segment granularity is recalculated. This process is repeated until the total inference chain length covers the entire price range within the step limit.

[0039] In a power trading simulation scenario, identifying incomplete price ranges involves traversing the distribution of price-electricity pairs.

[0040] Specifically, the generation count of each segment within the high-price range is iterated. When the generation count of a segment is zero, it indicates that the agent did not output any price-electricity pairs for that segment during the execution of the pricing task, and this segment is identified as a pricing incomplete range. The reason for the formation of the pricing incomplete range is that the agent consumed too many inference steps when processing the pricing task in the low-price range, causing the step quota to be exhausted before reaching the high-price range.

[0041] It should be noted that the criteria for determining a price incomplete interval include two elements: first, the number of segments generated is zero; second, the segment is located within the price interval after the step quota has been exhausted. Segments that meet both conditions are marked as price incomplete intervals, and their total number is counted as the basis for the subsequent reallocation of segment numbers.

[0042] In one embodiment, dynamic programming is an optimization method that decomposes a complex problem into overlapping subproblems and avoids redundant computation by storing the solutions to the subproblems. In this embodiment, the dynamic programming algorithm uses the upper limit of the number of steps in a single round of inference as a constraint and maximizes the number of covered price intervals as the objective. The specific implementation process is as follows: The entire price interval is evenly divided into n sub-intervals according to price (n is a preset value, such as 10). For each sub-interval, a state variable is defined as the number of segments that can be allocated to that interval. The state transition equation is dp[i][j]=max(dp[i-1][k]+benefit(jk)), where dp[i][j] represents the maximum coverage benefit of using j segments in the first i sub-intervals, k is the number of segments used in the first i-1 intervals, and benefit(m) is the coverage benefit of allocating m segments to the current interval (calculated as m multiplied by the interval length). The algorithm solves the problem step by step from the low-price interval to the high-price interval, calculates the coverage benefit of allocating different numbers of segments in each sub-interval, and selects the allocation scheme that maximizes the total coverage benefit. By tracing back the state transition path, we obtain the allocation table of the number of segments for each price range.

[0043] For example, redundant segments are defined based on a comparison between the current number of segments generated and the average number of segments generated in the interval. When the number of segments generated exceeds the average number of segments generated in that interval, the segment is identified as a redundant segment, indicating that the segment has obtained above-average inference resources under the current allocation scheme. Furthermore, the segment number transfer process involves reallocating a portion of the number of redundant segments to the quotation-deficient interval.

[0044] In one possible implementation, a dynamic programming algorithm calculates an upper limit on the number of transferable segments for each redundant segment. This upper limit is equal to the difference between the number of segments generated and the average number of segments generated. The algorithm selects a transfer scheme under constraints, ensuring that the incomplete price range has enough segments to initiate the bidding task, while guaranteeing that the low-price range retains basic bidding coverage. After the transfer is complete, the segment number allocation table for each price range is updated. Based on the segment number allocation table, the segment granularity is recalculated for each price range. The segment granularity is calculated by dividing the price span of the price range by the adjusted number of segments.

[0045] For example, if the price span of a certain price range is a preset price range, and the adjusted number of segments is several, then the segment granularity of that range is the quotient of the price span and the number of segments. The larger the segment granularity, the wider the price range covered by each segment within that range.

[0046] Understandably, the calculation of the total length of the inference chain involves summing the estimated inference steps for each price range. The estimated inference steps for each price range are equal to the number of segments in that range multiplied by the average inference steps for a single segment, which is calculated based on the execution records of historical pricing tasks.

[0047] In one embodiment, the iterative adjustment process employs a loop-based judgment mechanism. When the adjusted total length of the inference chain exceeds the upper limit of the number of steps in a single round of inference, the price range with the smallest segment granularity is selected from the segment number allocation table. The smallest segment granularity of this range means that its number of segments is excessive relative to the price span, leaving room for further compression. After reducing the number of segments in this range, the segment granularity and the total length of the inference chain are recalculated, and the above judgment and adjustment process is repeated until the total length of the inference chain does not exceed the upper limit of the number of steps. This iterative adjustment mechanism enables the segment number allocation scheme to cover the entire price range while satisfying the step constraint.

[0048] Step S105: Based on the adjusted segmentation granularity of each interval, reconfigure the generation order and step allocation of each price interval in the agent inference chain, obtain the updated price curve generation sequence, apply the sequence to generate price-electricity pairs in the agent decision, evaluate whether the adjusted generated price sequence has covered the price upper limit, and obtain the complete generated price curve.

[0049] Obtain the adjusted segmentation granularity and segmentation number allocation table for each price interval. Arrange the price intervals in ascending order of price. Allocate a corresponding inference step quota for each price interval. The inference step quota is calculated based on the number of segments in the interval and the average inference step count per segment, forming a generation order and step allocation configuration table for each price interval. This configuration table records the execution priority and available inference step resources for each price interval. Based on the generation order and step allocation configuration table, construct a price curve generation sequence. This sequence includes the segmentation boundaries, segmentation granularity, and corresponding step quotas for each price interval. During the agent decision-making process, execute the pricing task for each price interval sequentially according to the price curve generation sequence. The agent outputs price-electricity pairs within each segment and records the position of the price interval that has been output. Using the position of the price interval that has been output, determine whether the highest price of the current pricing sequence has reached the preset price limit. If the highest price has reached the price limit, it is determined that the pricing sequence has completed the coverage of the entire price interval, resulting in a fully generated pricing curve.

[0050] In the power trading simulation scenario, the construction of the generation order and step allocation configuration table is the core step in the reconfiguration of the agent inference chain.

[0051] Specifically, after obtaining the adjusted allocation table of segment granularity and number of segments for each price range, the price ranges are arranged in ascending order of price. This arrangement determines the processing order of the agent when executing the quotation task. For each price range, the inference step quota for that range is calculated based on its number of segments and the average inference steps per segment, and the calculation results are recorded in the configuration table.

[0052] It should be noted that the execution priority is determined by the position of the price range on the price axis. The execution priority of the lower price range is higher than that of the higher price range. This order setting is consistent with the business characteristics of the electricity market, where bids gradually increase from low to high.

[0053] In one embodiment, the construction process of the price curve generation sequence is as follows: Based on the generation order and step allocation configuration table, the segment boundaries, segment granularity, and corresponding step quotas of each price range are arranged according to execution priority to form the price curve generation sequence. Each element of this sequence corresponds to a price range's pricing task description, including the price range of the range, the number of segments, the price span of each segment, and the available inference step resources. During the decision-making process, the agent sequentially reads the task descriptions of each price range according to the order in the sequence and executes the price-electricity pair output operation within the corresponding segment.

[0054] For example, when performing a quotation task, the agent starts from the first element of the quotation curve generation sequence and outputs price-energy pairs sequentially according to the segment boundaries within the price range. After completing the output of each segment, the agent updates the record of the position of the completed price range. Further, the coverage assessment process uses the position of the completed price range for judgment. The upper price boundary corresponding to the highest price segment in the current quotation sequence is obtained and compared with a preset price upper limit. If the highest price has reached the price upper limit, it indicates that the agent has completed the entire quotation task from low to high price, and the quotation sequence covers the complete price range, thus obtaining a complete quotation curve containing price-energy pair data for each segment within the entire price range.

[0055] In some embodiments, the method for analyzing the decision-making behavior of electricity market trading agents further includes the following steps: The Monte Carlo simulation algorithm is used to evaluate the fitting accuracy and strategy exploration completeness of the quoted price curve. The results of the simulation are analyzed to improve the coverage of the high-price range. Feedback data on agency behavior in electricity market transactions are obtained to identify the improvement effect of the quoted price distortion in the feedback data. The segmented granularity adjustment parameters of the price range are iteratively updated to determine the inference resource allocation scheme.

[0056] A Monte Carlo simulation algorithm is used to randomly sample the fully generated price curve multiple times. In each sampling round, a price fluctuation scenario in the electricity market is simulated. The price curve is compared with the market clearing price under the simulated scenario, and the deviation between the price of each segment and the market clearing price is calculated to obtain the fitting accuracy index of the price curve. At the same time, the proportion of the number of times each price interval is covered by sampling is calculated to obtain the strategy exploration completeness index. Based on the fitting accuracy index and the strategy exploration completeness index, the coverage frequency of the high-price interval segment in multiple simulations is extracted. The current coverage rate is compared with the coverage rate before adjustment to obtain the coverage improvement result of the high-price interval. Simultaneously, the agent's bidding output behavior and corresponding market response data in each simulation round are recorded to form an agent behavior feedback dataset. The agent behavior feedback dataset is obtained, and segments in which the deviation between the bid price and the market clearing price exceeds a preset deviation threshold are identified. These segments are marked as bidding distortion segments, and the change in the number of bidding distortion segments before and after adjustment is calculated to obtain the improvement effect of bidding distortion. If the improvement effect does not reach the preset improvement threshold, the number of segments in the corresponding interval is increased according to the distribution position of the price distortion segments, the step quota ratio of the high-price interval is adjusted according to the coverage improvement result, the segment granularity adjustment parameters are iteratively updated, and the updated inference resource allocation scheme is obtained, so as to realize the physical realizable coverage of the agent's price range price generation under the constraint of the upper limit of inference steps.

[0057] In power trading simulation scenarios, the Monte Carlo simulation algorithm is a statistical method based on numerical estimation using extensive random sampling. This algorithm performs multiple rounds of random sampling on the price curve, constructing different market price fluctuation scenarios in each round, thereby statistically evaluating the performance of the price curve under various market conditions. The number of random sampling rounds is pre-set according to the required evaluation accuracy; the more sampling rounds, the higher the confidence level of the evaluation results.

[0058] Specifically, in each round of sampling, the Monte Carlo simulation algorithm randomly selects a set of market clearing prices as the simulation scenario for the current round, based on the price fluctuation distribution characteristics in historical electricity market data. These market clearing prices reflect the equilibrium price level formed in the electricity market under specific supply and demand conditions.

[0059] It should be noted that the calculation process for the fitting accuracy index is as follows: For each segment of the price curve, the price of that segment is compared with the market clearing price of the corresponding price range in the current simulation scenario, and the absolute deviation between the two is calculated. After calculating the absolute deviation values ​​for all segments, all deviation values ​​are statistically summarized, and the root mean square error is used as the fitting accuracy index for the entire price curve. The smaller the value of the fitting accuracy index, the higher the degree of fit between the price curve and the market clearing price, and the stronger the consistency between the agent's pricing behavior and the actual market price trend. After multiple rounds of sampling, the fitting accuracy index of each round is averaged to obtain the comprehensive fitting accuracy of the price curve under various market conditions.

[0060] In one embodiment, the strategy exploration completeness index is derived by statistically analyzing the proportion of times each price range is covered by sampling out of the total number of samplings. If a price range is covered in multiple rounds of sampling, the strategy exploration completeness for that range is high; if a price range is covered only in a few rounds, it indicates that the agent's price exploration for that range is insufficient.

[0061] For example, the coverage improvement result in the high-price range is obtained by comparing the number of coverages before and after the adjustment. The current coverage rate is obtained by dividing the number of times the current price curve covers the high-price range by the total number of samples; this coverage rate is then subtracted from the baseline coverage rate before the adjustment to obtain the coverage improvement value. Further, the process of forming the agent behavior feedback dataset includes: recording the price-electricity pair output by the agent, the corresponding market clearing price, and the matching status between the two in each round of simulation. The matching status indicates whether the agent's offer was accepted by the market in that round. The records from multiple rounds of simulation are aggregated to form a complete dataset containing the relationship between agent pricing behavior and market response.

[0062] In one possible implementation, the identification of pricing distortion segments is based on a preset deviation threshold. The deviation records of each segment in the agent behavior feedback dataset are traversed. If the average deviation of a segment in multiple rounds of simulation exceeds the preset deviation threshold, the segment is marked as a pricing distortion segment. The change in the number of pricing distortion segments before and after segment granularity adjustment is statistically analyzed. If the number of pricing distortion segments decreases after adjustment, it indicates that pricing distortion has been improved; the difference in the number before and after adjustment is used as a quantitative indicator of the improvement effect.

[0063] Understandably, the iterative update of the segment granularity adjustment parameters is performed based on the comparison between the improvement effect and the preset improvement threshold. If the improvement effect does not reach the preset improvement threshold, the distribution position of the price distortion segments on the price axis is analyzed, and the number of segments in the price range where the price distortion segments are concentrated is increased, making the segment granularity within that range finer, thereby improving the price accuracy. Simultaneously, based on the coverage improvement results of the high-price range, if the coverage of the high-price range is still lower than the preset coverage target, the proportion of the high-price range in the total step quota is increased, allowing the agent to obtain more inference step resources in the high-price range. Through the above adjustments, the updated segment granularity adjustment parameters are obtained, and the segment number allocation and inference step quota division are re-executed accordingly, forming an updated inference resource allocation scheme.

[0064] Preferably, when the improvement effect reaches the preset improvement threshold and the coverage rate of the high-price range reaches the preset coverage rate target, the iterative update process is stopped, confirming that the current inference resource allocation scheme has achieved the physically achievable coverage of the agent's bids for the entire price range under the constraint of the upper limit of inference steps.

[0065] Please refer to Figure 2 , Figure 2 This invention illustrates an analysis device for agent decision-making behavior in electricity market transactions, comprising: The data acquisition module is used to acquire the number of price-electricity pairs generated in each segment, the percentage of steps consumed in the current inference chain, and the actual coverage of the high-price segment during the agent decision-making process of the power trading simulation platform. The resource identification module is used to analyze the remaining available inference steps based on the proportion of steps consumed in the current inference chain. When the ratio of the remaining available inference steps to the actual coverage of the high-price interval is lower than a preset ratio threshold, the module obtains the inference resource shortage interval where the number of generated price-power pairs in the high-price interval is lower than the generation number threshold, and determines the segmentation granularity division scheme that needs to be adjusted based on the inference resource shortage interval. The task partitioning module is used to re-divide the generation tasks of each price range according to the segmentation granularity partitioning scheme that needs to be adjusted, under the constraint of the upper limit of the number of inference steps in a single round of the agent, and obtain the price-electricity pair generation distribution based on the generation task. The interval adjustment module is used to identify incomplete price intervals in the high-price intervals that have not yet been generated due to exhaustion of steps, based on the price-electricity pair generation distribution, reallocate the number of segments to the high-price intervals, and calculate the segment granularity of each interval after adjustment. The sequence configuration module is used to reconfigure the generation order and step allocation of each price interval in the agent inference chain based on the segmented granularity of each interval after adjustment, obtain the updated price curve generation sequence, apply the sequence to generate price-electricity pairs in the agent decision, evaluate whether the adjusted generated price sequence has covered the price upper limit, and obtain the complete generated price curve.

[0066] Furthermore, in some embodiments, the device may further include: The evaluation iteration module is used to evaluate the fitting accuracy and strategy exploration completeness of the price curve using the Monte Carlo simulation algorithm, analyze the coverage improvement results of the high-price range in the simulation results, obtain feedback data on agency behavior in electricity market transactions, identify the improvement effect of price distortion in the feedback data, and iteratively update the segmented granularity adjustment parameters of the price range to determine the inference resource allocation scheme.

[0067] The above description of the embodiments is only for the purpose of helping to understand the technical solutions and core ideas of this application; those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for analyzing agent decision-making behavior in electricity market transactions, characterized in that, The method includes: In the agent decision-making process of the power trading simulation platform, obtain the number of price-electricity pairs generated in each segment, the percentage of steps consumed in the current inference chain, and the actual coverage of the high-price range. The remaining available inference steps are analyzed based on the percentage of steps consumed in the current inference chain. When the ratio of the remaining available inference steps to the actual coverage of the high-price range is lower than a preset ratio threshold, the inference resource shortage range where the number of generated price-power pairs in the high-price range is lower than the generation number threshold is obtained. Based on the inference resource shortage range, the segmentation granularity division scheme that needs to be adjusted is determined. Based on the segmentation granularity partitioning scheme that needs to be adjusted, the generation tasks for each price range are re-divided under the constraint of the upper limit of the number of steps in a single round of agent inference, and the price-electricity pair generation distribution is obtained based on the generation task. Based on the price-electricity pair generation distribution, identify the incomplete quotation intervals in the high-price intervals that have not yet been generated due to exhaustion of steps, reallocate the number of segments to the high-price intervals, and calculate the segment granularity of each interval after adjustment. Based on the adjusted segmented granularity of each interval, the generation order and step allocation of each price interval in the agent inference chain are reconfigured to obtain the updated price curve generation sequence. This sequence is then applied in the agent decision to generate price-electricity pairs. The adjusted generated price sequence is evaluated to determine whether it has covered the upper limit of the price, thus obtaining the complete generated price curve.

2. The method for analyzing agent decision-making behavior in electricity market transactions according to claim 1, characterized in that, The process of obtaining the number of price-electricity pairs generated in each segment, the percentage of steps consumed in the current inference chain, and the actual coverage of the high-price segment during the agent decision-making process of the power trading simulation platform includes: The agent captures the price-energy pairs output in each price segment in real time, records the number of price-energy pairs generated in each segment, arranges them in order from low to high price axis to form a segment generation number sequence, and marks the price boundary value corresponding to each segment to obtain a segment generation statistics table. Based on the segmented statistical table, the number of inference steps executed in the current inference chain of the agent and the preset single-round inference step limit are read, and the percentage of steps consumed in the current inference chain is calculated. When the percentage of consumed steps exceeds the step percentage threshold and the number of generated high-price intervals is lower than the number of generated intervals threshold, the high-price interval is marked as an under-covered interval. Based on the location identifiers of the under-covered intervals, coverage statistics are performed on the high-price intervals to obtain the actual coverage rate of the high-price intervals.

3. The method for analyzing agent decision-making behavior in electricity market transactions according to claim 1, characterized in that, The remaining available inference steps are analyzed based on the percentage of steps consumed in the current inference chain. When the ratio of the remaining available inference steps to the actual coverage of the high-price range is lower than a preset ratio threshold, an inference resource shortage range is identified where the number of price-power pairs generated in the high-price range is lower than a generation number threshold. Based on the inference resource shortage range, the segmentation granularity scheme to be adjusted is determined, including: Obtain the remaining available inference steps in the agent's current inference chain and the actual coverage of the high-price range. Divide the remaining available inference steps by the actual coverage to obtain the step coverage ratio. When the step coverage ratio is lower than the preset ratio threshold, the high-price interval is marked as an inference resource shortage interval, and the starting price boundary and ending price boundary of the inference resource shortage interval on the price axis are recorded. The price span of the inference resource shortage interval is determined based on the starting price boundary and the ending price boundary on the price axis. The current segment granularity is calculated based on the price span of the inference resource shortage interval and the number of currently allocated segments. Based on the comparison between the remaining available inference steps and the estimated number of steps required to complete full coverage of the interval, the degree of resource gap is determined, and the segment boundaries of the interval with insufficient inference resources are re-divided according to the degree of resource gap, resulting in an adjusted segmentation granularity scheme.

4. The method for analyzing agent decision-making behavior in electricity market transactions according to claim 1, characterized in that, The process of re-dividing the generation tasks for each price range according to the adjusted segmentation granularity scheme, under the constraint of the upper limit of the number of inference steps in a single round of proxy, and obtaining the price-electricity pair generation distribution based on the generation tasks includes: Extract the adjusted number of segments for each price range from the segmentation granularity division scheme that needs adjustment, and simultaneously read the maximum number of steps for a single round of agent inference; Arrange the price ranges in ascending order of price to form a price range sorting sequence; For each price range, the adjusted number of segments is used to divide the maximum number of steps in a single round of inference according to the proportion of the number of segments in each range to the total number of segments, thus obtaining the step quota table for each price range. Based on the step quota table for each price range, the task of generating price-energy pairs is executed sequentially according to the segment boundaries within each price range. After the price-energy pairs for each segment are output, the number of generated pairs for that segment is recorded. The distribution of generated price-energy pairs is obtained by summarizing them according to the price axis position.

5. The method for analyzing agent decision-making behavior in electricity market transactions according to claim 1, characterized in that, The generation order and step allocation of each price interval in the agent inference chain are reconfigured based on the segmented granularity of each interval to obtain the updated price curve generation sequence. This sequence is applied in the agent decision-making process to generate price-electricity pairs. The process evaluates whether the adjusted generated price sequence has covered the price upper limit, resulting in a complete generated price curve, including: Obtain the segmentation granularity and segmentation number allocation table for each interval after adjustment, and arrange each price interval in order of price from low to high; For each price range, a corresponding inference step quota is allocated. The inference step quota is calculated based on the number of segments in the range and the average number of inference steps per segment, forming a generation order and step allocation configuration table for each price range. Construct a quotation curve generation sequence based on the generation order and step allocation configuration table; During the agent decision-making process, the agent executes the pricing task for each price range sequentially according to the pricing curve generation sequence. The agent outputs price-electricity pairs in each segment and records the position of the price range that has been output. Determine whether the highest price in the current price sequence has reached the preset price limit. If it has, then the price sequence has completed the coverage of the entire price range, resulting in a complete price curve.

6. The method for analyzing agent decision-making behavior in electricity market transactions according to claim 5, characterized in that, The price quote curve generation sequence includes the segment boundaries, segment granularity, and corresponding step quotas for each price range.

7. The method for analyzing agent decision-making behavior in electricity market transactions according to claim 1, characterized in that, The method further includes: The Monte Carlo simulation algorithm is used to evaluate the fitting accuracy and strategy exploration completeness of the quoted price curve. The results of the simulation are analyzed to improve the coverage of the high-price range. Feedback data on agency behavior in electricity market transactions are obtained to identify the improvement effect of the quoted price distortion in the feedback data. The segmented granularity adjustment parameters of the price range are iteratively updated to determine the inference resource allocation scheme.

8. The method for analyzing agent decision-making behavior in electricity market transactions according to claim 7, characterized in that, The Monte Carlo simulation algorithm is used to evaluate the fitting accuracy and strategy exploration completeness of the price curve, analyze the coverage improvement results of the high-price range in the simulation results, obtain feedback data on agency behavior in electricity market transactions, identify the improvement effect of price distortion in the feedback data, and iteratively update the segmented granularity adjustment parameters of the price range to determine the inference resource allocation scheme, including: The Monte Carlo simulation algorithm is used to perform multiple rounds of random sampling on the fully generated price curve, simulating price fluctuation scenarios in the electricity market in each round of sampling; The price curve is compared with the market clearing price in the simulated scenario, and the deviation between the price of each segment and the market clearing price is calculated to obtain the fitting accuracy index of the price curve. The proportion of times each price range was covered by sampling out of the total number of samplings was used to obtain the strategy exploration completeness index; Based on the fitting accuracy index and the strategy exploration completeness index, the number of times the high-price interval is covered in multiple rounds of simulation is extracted. The current coverage rate is compared with the coverage rate before adjustment to obtain the coverage improvement result of the high-price interval. Record the agent's pricing output behavior and corresponding market response data in each simulation round to form an agent behavior feedback dataset; From the agent behavior feedback dataset, identify segments where the deviation between the quoted price and the market clearing price exceeds a preset deviation threshold, mark them as price distortion segments, and count the change in the number of price distortion segments before and after adjustment to obtain the improvement effect of price distortion. When the improvement effect does not reach the preset improvement threshold, the number of segments in the corresponding interval is increased according to the distribution position of the price distortion segments, and the step quota ratio of the high-price interval is adjusted according to the coverage improvement result. The segment granularity adjustment parameters are iteratively updated to obtain the inference resource allocation scheme.

9. An analysis device for agent decision-making behavior in electricity market transactions, characterized in that, The device includes: The data acquisition module is used to acquire the number of price-electricity pairs generated in each segment, the percentage of steps consumed in the current inference chain, and the actual coverage of the high-price segment during the agent decision-making process of the power trading simulation platform. The resource identification module is used to analyze the remaining available inference steps based on the proportion of steps consumed in the current inference chain. When the ratio of the remaining available inference steps to the actual coverage of the high-price interval is lower than a preset ratio threshold, the module obtains the inference resource shortage interval where the number of generated price-power pairs in the high-price interval is lower than the generation number threshold, and determines the segmentation granularity division scheme that needs to be adjusted based on the inference resource shortage interval. The task partitioning module is used to re-divide the generation tasks of each price range according to the segmentation granularity partitioning scheme that needs to be adjusted, under the constraint of the upper limit of the number of inference steps in a single round of the agent, and obtain the price-electricity pair generation distribution based on the generation task. The interval adjustment module is used to identify incomplete price intervals in the high-price intervals that have not yet been generated due to exhaustion of steps, based on the price-electricity pair generation distribution, reallocate the number of segments to the high-price intervals, and calculate the segment granularity of each interval after adjustment. The sequence configuration module is used to reconfigure the generation order and step allocation of each price interval in the agent inference chain based on the segmented granularity of each interval after adjustment, obtain the updated price curve generation sequence, apply the sequence to generate price-electricity pairs in the agent decision, evaluate whether the adjusted generated price sequence has covered the price upper limit, and obtain the complete generated price curve.

10. The analysis device for agent decision-making behavior in electricity market transactions as described in claim 9, characterized in that, The device further includes: The evaluation iteration module is used to evaluate the fitting accuracy and strategy exploration completeness of the price curve using the Monte Carlo simulation algorithm, analyze the coverage improvement results of the high-price range in the simulation results, obtain feedback data on agency behavior in electricity market transactions, identify the improvement effect of price distortion in the feedback data, and iteratively update the segmented granularity adjustment parameters of the price range to determine the inference resource allocation scheme.

Images