A Virtual Power Plant Generation-Consumption-Pricing Collaborative Optimization System Based on AI Large Model

Abstract

This invention relates to the field of collaborative optimization technology, specifically to a virtual power plant generation-consumption-price collaborative optimization system based on an AI large-scale model. The system includes a load confidence matching module, a resource stability mapping module, a source-load capacity coupling module, a price range adjustment module, and a comprehensive control linkage module. This invention uses a hybrid neural network model based on load behavior data and equipment temperature control characteristic sequences to predict the confidence range of load demand. The overlap between the predicted range and the generation response characteristics is used to measure the dispatch matching confidence. A stability screening mechanism for regulating resources is constructed by combining output volatility and equipment inertia characteristics, improving the controllability of load regulation and the real-time performance of source-side response. The price adjustment rhythm is corrected through a price response delay factor, achieving dynamic closed-loop linkage between load regulation and price guidance. The execution priority is dynamically updated under multiple response condition matching conditions, improving the determinism of resource scheduling and the sensitivity of response.

Description

Technical Field

[0001] This invention relates to the field of collaborative optimization technology, and in particular to a virtual power plant generation-consumption-price collaborative optimization system based on an AI large model. Background Technology

[0002] The field of collaborative optimization technology involves modeling the coupling relationships between multiple systems or subsystems, constructing and solving optimization problems, and aiming to achieve globally optimal configuration among multiple links and objectives. In complex systems such as power systems, industrial processes, and integrated energy management, subsystems have mutual influences and constraints. Examples include resource allocation and scheduling coordination between the power supply side, load side, and energy storage side, as well as collaborative optimization among multiple objectives such as economy, safety, and stability. This field often combines large-scale linear / nonlinear programming, mixed-integer programming, multi-objective optimization, and distributed optimization algorithms. Through precise modeling and iterative solution strategies, it promotes the improvement of overall system performance and enhances resource utilization efficiency and operational stability.

[0003] The Virtual Power Plant Generation-Consumption-Pricing Coordinated Optimization System refers to a unified, coordinated, and optimized dispatching system built to address the interrelationships between the power supply side, load side, and electricity pricing mechanism within a virtual power plant. This system achieves global optimized dispatching of virtual power plant resources by coordinating power generation planning, user load regulation, and dynamic pricing strategies. Its purpose is to enhance the economic benefits and dispatching flexibility of virtual power plants in market participation, while strengthening their responsiveness to grid frequency regulation and peak shaving demands. It is suitable for scenarios such as demand response execution, load aggregation management, and electricity spot market trading.

[0004] Traditional optimization systems rely on static scheduling rules to execute load regulation and power generation coordination, lacking dynamic identification methods for changing trends in user electricity consumption behavior. In situations where load-side responses exhibit nonlinear changes or frequent fluctuations, insufficient scheduling matching or response lag issues are prone to occur. Electricity price mechanism adjustments do not fully consider the lag effect of delayed response on load changes, which can easily lead to the pace of electricity price adjustments lagging behind actual load changes. This can cause price incentive failure or resource allocation deviations. For example, during periods of frequent peak load changes, a slow pace of electricity price adjustments will cause the load-side response to lag behind dispatch instructions, reducing market revenue and regulation efficiency. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a virtual power plant generation-consumption-price collaborative optimization system based on an AI large model.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a virtual power plant generation-consumption-price collaborative optimization system based on an AI large-scale model, the system comprising: The load confidence matching module, based on the target virtual power plant area, calls a hybrid model of convolutional neural network-long short-term memory network pre-trained based on load clustering labels and temperature control condition feature sequences. It performs boundary overlap rate calculation on the predicted load confidence interval and the power generation output response range to generate large model prediction mapping matching information. The resource stability mapping module extracts the output fluctuation rate, cold load recovery rate and equipment thermal inertia parameters of three consecutive control cycles from the set of power generation units with overlapping boundaries in the large model prediction mapping matching information, and filters the resource set that meets the stability change threshold, and generates a power generation resource stability screening set. The source-load capacity coupling module calls the output adjustment range of the power generation resource stability screening set, extracts the load change slope of the load-side equipment, compares the power slope of the source and the load, extracts the coupling terms that meet the slope difference tolerance threshold, and generates a source-load slope coupling list. The electricity price range adjustment module calls the set of loads that have not been matched in the source-load slope coupling list, calculates the lag influence factor, applies a step size adjustment to the upper and lower limits of the current electricity price range, and generates dynamic adjustment parameters for the electricity price range.

[0007] The present invention improves upon the following: the large model prediction mapping matching information includes the predicted load confidence boundary distribution, the output response tolerance range, and the overlap rate compliance label; the power generation resource stability screening set includes the regulation cycle stability score, the historical thermal inertia variation amplitude, and the output fluctuation threshold verification label; the source-load slope coupling list includes the response adjustment comparison table, the slope matching level label, and the cycle matching validity record; and the electricity price range dynamic adjustment parameters specifically include the step size adjustment coefficient, the price upper and lower limit variation amplitude, and the regulation cycle range identifier.

[0008] The present invention is improved in that the load confidence matching module includes: The behavior data extraction submodule collects the active power time series, start-stop frequency and temperature control setpoint on the user side based on the target virtual power plant area, divides them into labels according to equipment category, and assigns them to cluster identifier sequences and temperature control feature sequences respectively. The cluster identifier and temperature control sequence are then bound to form a behavior sample vector set to generate a load behavior sample set. The load confidence prediction submodule, based on the load behavior sample set, calls the trained convolutional neural network-long short-term memory network model, combines the target prediction period, outputs the upper and lower confidence limits of each type of equipment within the period, establishes the confidence interval boundary sequence of the predicted load, records the corresponding time step of the prediction curve, and generates a confidence boundary value sequence. The boundary matching judgment submodule collects the standard deviation coefficient, cold start time and maximum ramp rate of the power generation unit based on the confidence boundary value sequence, constructs the upper and lower boundaries of the output response of each power generation unit, calculates the interval overlap rate between the predicted load confidence boundary and the output response boundary, and compares the overlap rate with the confidence overlap threshold to generate large model prediction mapping matching information.

[0009] The present invention is improved in that the resource stabilization mapping module includes: The parameter extraction submodule marks the set of power generation units with overlapping boundaries in the large model prediction mapping matching information, collects the output records of each power generation unit in three consecutive control cycles, extracts the output data sequence, cold load response curve and thermal parameter change records during equipment operation, and calculates the output fluctuation rate, cold load recovery rate and equipment thermal inertia parameters respectively to obtain the set of periodic operation parameters. The fluctuation analysis submodule calls the output fluctuation rate, cold load recovery rate and equipment thermal inertia parameters in the set of periodic operation parameters, establishes a sliding window sequence according to time steps in three control cycles, calculates the sliding standard deviation corresponding to each parameter, takes the difference between the maximum and minimum values ​​of the standard deviation sequence as the fluctuation range, calculates the total resource fluctuation sequence, labels the resources according to the resource fluctuation sequence, and generates resource fluctuation label values. The resource screening submodule determines whether the volatility of each power generation unit is lower than the set resource stability change threshold based on the resource volatility label value, filters out resource units that meet the conditions, and records the control cycle number, stability identifier and volatility level to obtain a stable power generation resource screening set.

[0010] The present invention is improved in that the source-load capacity coupling module includes: The output parameter extraction submodule calls the set of power generation units in the stable screening set of power generation resources, extracts the minimum and maximum output values ​​of the power generation units in the current scheduling cycle, calculates the output change amplitude per unit time within the interval, and normalizes it as the source-side output adjustment range. At the same time, it collects the load response rate during the equipment operation period, obtains the power change rate data sequence of the load-side equipment, and obtains the source-load output change sequence. The slope construction submodule classifies and pairs the source-side output adjustment range with the load-side power change rate according to the source-load output change sequence, constructs source-load power time series respectively, extracts slope change points in each time period and converts them into a difference sequence under a unified time scale, constructs source-load power change slope difference sequence periodically, and obtains source-load power slope difference value sequence. The slope matching and filtering submodule calculates the average difference and fluctuation deviation rate within the scheduling cycle for each set of data in the source-load power slope difference sequence, and constructs a coupling adaptation index in combination with the response hysteresis time factor of the power generation unit. It then calculates and obtains the coupling adaptation index value sequence of the source-load pair, filters the source-load pairs whose index values ​​are lower than the source-load coupling tolerance threshold, establishes a coupling validity mark, and obtains the source-load slope coupling list.

[0011] The present invention is improved in that the electricity price range adjustment module includes: The delayed data extraction submodule calls the set of loads that have not been matched in the source-load slope coupling list, collects the unit power change of the load in the current scheduling cycle and the previous cycle, obtains the start time and arrival time of the output offset in the load response curve, calculates the adjustment range and response delay of each load device, establishes the correspondence between load number and adjustment-delay, and obtains the adjustment delay matching sequence. The impact factor calculation submodule collects the real-time total load value and the predicted load value within the scheduling period according to the adjustment delay matching sequence, calculates the difference between the two as the load offset, extracts the average response delay value of the current load set, the maximum load adjustment range and the average change of the load value within the current period, and calculates the lag impact factor. The price boundary correction submodule determines whether the lag factor is higher than the set electricity price change trigger ratio threshold based on the lag factor. If the condition is met, it extracts the upper and lower limits of the price range in the previous period and the minimum and maximum response electricity values ​​in the current period, calculates the adjustment step size according to the floating ratio of the upper and lower limit boundaries, and performs addition and subtraction operations on the upper and lower limits of the range in sequence to obtain the dynamic adjustment parameters of the electricity price range.

[0012] The present invention has an improvement, wherein the system further includes: The integrated control and linkage module calls the response delay and price elasticity coefficient of the matching source load items in the power price range dynamic adjustment parameters and source load slope coupling list to construct the correspondence matrix between the control window and the price sensitive section. Based on whether it falls into the window execution section on the current time axis, it updates the priority execution order and generates the control execution priority list. The priority list for regulation execution includes source-load response ranking numbers, electricity price sensitivity weight values, and regulation window segment labels.

[0013] The present invention is improved in that the integrated control and linkage module includes: The response parameter collection submodule calls the dynamically adjusted parameters of the electricity price range and the matched source load items in the source load slope coupling list, extracts the response delay and electricity price elasticity coefficient of each source load pair, classifies and marks them according to the scheduling time axis, and constructs the time-series response array of the source load pair in the current period to obtain the source load time-series response index set. The adjustment window construction submodule extracts the response delay value and the electricity price elasticity coefficient value based on the source-load time sequence response index set, calculates the intersection period with the current electricity price adjustment range, maps it to the time step in the time axis, determines whether each source-load pair falls into the intersection segment, marks it as adjustable, and generates a source-load adjustment effective segment table. The priority update submodule extracts the response delay and electricity price elasticity level values ​​based on the source-load pairs marked as adjustable in the effective range table of source-load regulation. According to the synchronicity and sensitivity level of the two in the regulation range, it calculates the priority index value and reorders the adjustable source-load items in the current cycle from high to low according to the index value to obtain the regulation execution priority list.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, a hybrid neural network model based on load behavior data and equipment temperature control characteristic sequences is used to predict the confidence interval of load demand. The overlap between the predicted interval and the boundary of the power generation response characteristics is used to measure the confidence of scheduling matching. A stability screening mechanism for regulating resources is constructed by combining output volatility and equipment inertia characteristics. The controllability of load regulation and the real-time performance of source-side response are improved through the coupling matching process of power change slope between source and load. The price adjustment rhythm is corrected by the electricity price response delay factor to achieve dynamic closed-loop linkage between load regulation and price guidance. The execution priority is dynamically updated under multiple response conditions, which improves the determinism of resource scheduling and the sensitivity of response, and enhances the system's rhythm consistency and revenue optimization capabilities in multi-cycle coordination. Attached Figure Description

[0015] Figure 1 This is a system flowchart of the present invention; Figure 2 This is a flowchart of the load confidence matching module of the present invention; Figure 3 This is a flowchart of the resource stabilization mapping module of the present invention; Figure 4 This is a flowchart of the source-load capacity coupling module of the present invention; Figure 5 This is a flowchart of the electricity price range adjustment module of the present invention; Figure 6 This is a flowchart of the integrated control and linkage module of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0017] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0018] Please see Figure 1 The present invention provides a technical solution: a virtual power plant generation-consumption-price collaborative optimization system based on an AI large model, the system including a load confidence matching module, a resource stability mapping module, a source-load capacity coupling module, a price range adjustment module, and a comprehensive control linkage module; The load confidence matching module, based on the target virtual power plant area, collects historical behavior data composed of the active power time series, start-stop frequency, and temperature control setpoint of each type of user-side electrical equipment. It calls a hybrid model of convolutional neural network-long short-term memory network pre-trained based on load clustering labels and temperature control condition feature sequences to predict the confidence interval of the load for the target period, obtains the upper and lower confidence limits of the predicted load, collects the standard deviation coefficient, cold start time, and maximum ramp rate of the power generation unit, performs boundary overlap rate calculation on the predicted load confidence interval and the power generation output response range, determines whether the boundary overlap is less than the set confidence overlap threshold, and generates large model prediction mapping matching information. The resource stability mapping module extracts the output fluctuation rate, cold load recovery rate and equipment thermal inertia parameters for three consecutive control cycles from the set of power generation units with overlapping boundaries in the large model prediction mapping matching information. It calculates the sliding standard deviation and period variation amplitude of the three data respectively, screens the resource set that meets the stability variation threshold, and generates a power generation resource stability screening set. The source-load capacity coupling module calls the output adjustment range of the power generation resource stability screening set, extracts the load change slope of the load-side equipment, constructs the power change rate curve of the source-load pair within the scheduling cycle, compares the power slope of the source and the load, extracts the coupling terms that meet the slope difference tolerance threshold, and generates the source-load slope coupling list. The electricity price range adjustment module calls the set of loads that have not been matched in the source-load slope coupling list, collects the load adjustment range and output response delay in the current and previous cycles, calculates the lag effect factor between the delay and the real-time total load difference, compares the lag effect factor with the set electricity price change trigger ratio threshold, and if it exceeds the threshold, applies a change step size ratio adjustment to the upper and lower limits of the current electricity price range to generate dynamic adjustment parameters for the electricity price range. The integrated control and linkage module calls the response delay and price elasticity coefficient of the matching source load items in the power price range dynamic adjustment parameters and source load slope coupling list to construct the correspondence matrix between the control window and the price sensitive section. Based on whether it falls into the window execution section on the current time axis, it filters the allowed control items, updates the priority execution order, and generates the control execution priority list. The large model prediction mapping matching information includes the distribution of predicted load confidence boundaries, output response tolerance range, and overlap rate compliance labels. The power generation resource stability screening set includes the regulation cycle stability score, historical thermal inertia variation amplitude, and output fluctuation threshold verification labels. The source-load slope coupling list includes a response regulation comparison table, slope matching level labels, and cycle matching validity records. The electricity price range dynamic adjustment parameters specifically include the step size adjustment coefficient, price upper and lower limit variation amplitude, and regulation cycle range identifier. The regulation execution priority list includes the source-load response ranking number, electricity price sensitivity weight value, and regulation window segment label.

[0019] Please see Figure 2 The load confidence matching module includes: The behavior data extraction submodule collects the active power time series, start-stop frequency and temperature control setpoint on the user side based on the target virtual power plant area, divides them into labels according to equipment category, and assigns them to cluster identifier sequences and temperature control feature sequences respectively. The cluster identifier and temperature control sequence are then bound to form a behavior sample vector set to generate a load behavior sample set. The behavioral data extraction submodule, based on the target virtual power plant area, first identifies user nodes within the selected area that include four main types of electrical equipment: air conditioners, refrigerators, electric water heaters, and washing machines. The data collection cycle for the smart meters corresponding to each device is set to 15 minutes, covering a continuous 30-day timeline. After collecting the time series data for air conditioner equipment, a 96-dimensional vector is constructed daily. The average active power is obtained by averaging the data from 0:00 to 24:00 each day. The standard deviation is calculated as follows: Statistically analyze the magnitude of power fluctuations between adjacent time points. If the power increase or decrease is greater than... This counts as one start-stop cycle, and the cumulative start-stop frequency is approximately twice per hour. The air conditioner's temperature control setpoint is set to... The refrigerator is Electric water heater is The washing machine is a non-temperature-controlled type, so this value is not set. Clustering labels are assigned to the aforementioned devices. to With temperature control feature number to Construct a behavior sample vector for each device, in the following format: The sample vectors are integrated to generate a load behavior sample set, which is used for subsequent deep time series learning modeling.

[0020] Table 1 Sample Table of User-Side Device Behavior Data

[0021] As shown in Table 1, the sample set data comes from the power and behavior information collection results of each device within the electricity meter statistical period, in order to ensure that the behavior feature sequence has structural consistency before subsequent model processing.

[0022] The load confidence prediction submodule, based on the load behavior sample set, calls the pre-trained convolutional neural network-long short-term memory network model, combines the target prediction period, outputs the upper and lower confidence limits of each type of equipment within the period, establishes the confidence interval boundary sequence of the predicted load, records the corresponding time step of the prediction curve, and generates a confidence boundary value sequence. After constructing the load behavior sample set, the load confidence prediction submodule calls the clustering identifier and temperature control features of each type of equipment in the sample set as structural input, and constructs a daily behavior subsequence based on 96 points. In the scenario where the prediction target is the next hour, the system extracts data from the past 96 hours to form a training sliding window. The system maps the input features into a two-dimensional matrix according to "each type of equipment × time point sequence", inputs the structure row by row and establishes the mapping relationship with the time point label. Then, confidence prediction intervals are generated at each 15-minute point in the prediction period, and the predicted power value is generated at each point. Calculate its standard deviation The system sets boundaries according to confidence intervals, and calculates boundary values ​​using a two-sided 95% confidence level. The value format is as follows: ; ; For example: In the first time step of prediction, the predicted mean of the air conditioning equipment is... Standard deviation Substituting into the above equation, we get: ; ; The corresponding generated confidence boundary sequence is This sequence is used to further determine the overlap rate with the power generation capacity boundary. The "2" used in the confidence boundary is a set boundary weight coefficient, which is estimated based on the location of 95% of the distribution points in the past samples. After quantitative analysis of different standard deviations in the experiment, it was confirmed that the misjudgment rate remained within 3% when "2" was set, and it maintained stability for equipment groups with a prediction sample capacity of more than 300 samples. Therefore, it was set as a fixed value and was not adjusted with time or sample type.

[0023] The boundary matching judgment submodule collects the standard deviation coefficient, cold start time and maximum ramp rate of the power generation unit based on the confidence boundary value sequence, constructs the upper and lower boundaries of the output response of each power generation unit, calculates the interval overlap rate between the predicted load confidence boundary and the output response boundary, and compares the overlap rate with the confidence overlap threshold to generate large model prediction mapping matching information. After obtaining the predicted confidence boundary, the boundary matching judgment submodule reads the performance parameters of each registered power generation unit item by item. The maximum ramp rate of a certain gas turbine to be matched with the air conditioning load is... The cold start time is The standard deviation coefficient is set to Based on the power value for the current forecast period Construct an upper bound for the response The lower bound is That is, the response range is , and the prediction interval Perform overlap determination; the overlap interval is... The overlap length is The prediction interval length is the same. The overlap rate is: ; The system sets the overlap rate threshold. The judgment condition is: if the overlap rate If it is, then it is determined to be a match; this threshold The setting is based on regression statistics of the matching rate curves between 10 generator sets and load types, comparing the overlap rate in... to After analyzing the number of successful responses and response error deviation rate for each value within the range in actual scheduling, it was found that when the overlap rate... When the error rate drops below 10% and the idle rate of scheduled resources is below 5%, this value is set as a fixed threshold and locked in actual deployment without adaptive adjustment, because it is not significantly affected by changes in load type.

[0024] Please see Figure 3 The resource stabilization mapping module includes: The parameter extraction submodule marks the set of power generation units with overlapping boundaries in the large model prediction mapping matching information, collects the output records of each power generation unit in three consecutive control cycles, extracts the output data sequence, cold load response curve and thermal parameter change records during equipment operation, and calculates the output fluctuation rate, cold load recovery rate and equipment thermal inertia parameters respectively to obtain the set of periodic operation parameters. After identifying sets of power generation units with overlapping boundaries, the parameter extraction submodule uses the marked unit ID as an index to batch collect the operating data of each power generation unit over three consecutive control cycles. Each cycle is 15 minutes long, with a sampling frequency of 1 time / minute, meaning each cycle contains 15 sampling points, resulting in a total of 45 output record sequences for each device. These output records are then structured into a three-dimensional array in chronological order, divided into three segments according to the cycle. Within each cycle, the average output sequence and volatility are calculated using a time-series sliding window approach. Finally, the output change rate is extracted for the 30 minutes following the breakpoint of each cycle. A cold load response curve is generated. The maximum recovery slope within the first 5 minutes and the stable range amplitude within the next 30 minutes are extracted from the curve to form a cold load recovery rate index. Curvature is extracted from the temperature monitoring records of each power generation unit during operation. The curvature of the thermal inertia performance curve is calculated using a mixture of the first derivative of temperature change and the second derivative of time, respectively, to calculate the degree of gradual change in the heating and cooling processes. After calculation, the above three results—output fluctuation rate, cold load recovery rate, and equipment thermal inertia—are generated into floating parameter records on a periodic basis. The results are normalized by interval and stored in the periodic operation parameter set. The interval normalization uses the following formula: ; Taking power output fluctuation rate as an example, the standard deviation of a certain unit G1 in the first period is: The minimum value of the whole sample is The maximum value is Then the normalization value is: ; This method processes the three types of parameters to construct a normalized set of periodic operating parameters, providing the input fields required for subsequent resource volatility analysis.

[0025] The fluctuation analysis submodule calls upon the output fluctuation rate, cold load recovery rate, and equipment thermal inertia parameters from the cycle operation parameter set. It establishes a sliding window sequence over three control cycles, calculating the sliding standard deviation for each parameter. The difference between the maximum and minimum values ​​of the standard deviation sequence is taken as the fluctuation amplitude, using the formula: ; The total resource volatility sequence is obtained through calculation. Based on the resource volatility sequence, the resources are labeled and generated to produce resource volatility label values. in, and These represent the normalized maximum and minimum values ​​of the active power output fluctuation rate of the power generation unit within the three control cycles, respectively. The data source is the standard deviation of the active power output time series after interval normalization. and These represent the normalized maximum and minimum values ​​of the cold load recovery rate over three periods, respectively. The values ​​are derived from the normalized standard deviation of the load recovery trend curve after a power outage during the response period. and These represent the normalized maximum and minimum sliding standard deviations of the equipment's thermal inertia parameters, respectively. The data source is the curvature of the temperature-time curve, extracted and normalized. This represents the average frequency fluctuation value of the power output of the power generation unit. It is extracted by dividing the number of inflection points on the power output curve per unit time by the duration of that period. The average ramp-up rate of the power generation unit over three cycles refers to the average percentage increase in output per unit time that exceeds the ramp-up limit. This represents the total fluctuation value of a single power generation unit in the current cycle. The resource stability change threshold is set based on the system's accumulated experience value of the allowable fluctuation range in coordinated scheduling. This threshold is set by the virtual power plant operation platform after performing a 95% confidence interval boundary regression fitting based on the statistical distribution of resource fluctuations in multiple cycles. It does not have a fixed constant value and is dynamically adjusted differently according to different regions. After calling the cycle running parameter set, the fluctuation analysis submodule sequentially reads the output fluctuation rate, cold load recovery rate, and equipment thermal inertia parameters for each power generation unit. For each parameter, it constructs a sliding window sequence of length 5, performs standard deviation calculation on the set of values ​​within the window to obtain a sliding standard deviation sequence, and extracts the maximum value from this sequence. and minimum value The difference is the periodic variation amplitude of this term. After processing the three types of parameters in sequence, the result is substituted into the following structured calculation formula: ; In this formula, the first term on the left-hand side represents the range of variation of the standard deviation of the normalized active power output volatility, and its numerator... This reflects the degree of inconsistency between the maximum and minimum output fluctuations of the power generation unit within three control cycles. A larger difference indicates more severe fluctuations. The denominator... The frequency jitter adjustment factor of the output curve is taken as the natural logarithm to suppress the nonlinear effects of frequently fluctuating equipment. By modulating the high-frequency changes into a "high-risk amplification" weighting method, the weight of frequently fluctuating equipment is increased in the fluctuation evaluation. Second item The sliding standard deviation of the cold load recovery rate is square-root compressed and included in the total volatility. This operation moderately increases the contribution of cold start recovery uncertainty to the overall volatility without abruptly increasing it, thus balancing the impact assessment of response lag. Third item This represents the degree of disturbance to stability caused by changes in thermal inertia, where the denominator... This is an adjustment term for the average over-limit ramp rate of the equipment (the percentage of output exceeding the ramp rate per unit time). The larger the value, the more aggressive the device is in the scheduling process. The formula adjusts this by increasing the value in the denominator. The weighting reduces the impact of thermal inertia fluctuations on equipment with high climbing frequency, reflecting a bias towards stable equipment.

[0026] The advantage of this formula lies in its ability to establish a comprehensive volatility index across the time domain by jointly controlling resource parameters from three different dimensions. The first term focuses on output response behavior, the second evaluates load recovery stability, and the third constrains the degree of hysteresis disturbance caused by thermal inertia during operation. The overall index... The smaller the value, the more stable the resource operation and the higher the prediction matching degree; the larger the value, the more uncertain the resource behavior in a continuous period, making it difficult to predict its response result after scheduling.

[0027] Taking power generation unit G1 as an example, substitute the following parameter values: , , ; , ; , ; ; Substituting, we get: ; ; ; This result indicates that the comprehensive volatility evaluation value of power generation unit G1 is... Set thresholds on the platform It falls under the category of stable resources, and its output, recovery, and thermal inertia have not shown any significant abnormal fluctuations. It can be given priority scheduling as a stable supply unit in the source-load matching resource pool for the next stage.

[0028] The resource screening submodule determines whether the volatility of each power generation unit is lower than the set resource stability change threshold based on the resource volatility label value, screens the resource units that meet the conditions, and records the control cycle number, stability identifier and volatility level to obtain the power generation resource stability screening set. The resource filtering submodule performs individual judgments on the calculated resource fluctuation label values ​​for each device. The value will be related to the resource stability change threshold set by the platform. Compare and set thresholds. Instead of being a constant, the platform constructs a density distribution sequence of all volatility data over the past 30 days based on factors such as different region types, unit structures, and load types. It then extracts the upper bound of the 95% confidence interval and generates it using a regression boundary model. The system ultimately sets the volatility threshold for region A within this period as follows: Then when a certain power generation unit's Time is considered a stable resource, such as G1 mentioned above. If G1 is selected as a stable unit, its control cycle number is marked as the group number, the stability identifier is recorded as "1", and the fluctuation level is "low". If the equipment RΔ value is close to the boundary but does not exceed the limit, the level is marked as "medium". Equipment that exceeds the threshold is removed. Finally, a stable set of power generation resources is generated for subsequent source-load coupling assessment.

[0029] Please see Figure 4 The source-load coupling module includes: The output parameter extraction submodule calls the set of power generation units in the stable power generation resource screening set, extracts the minimum and maximum output values ​​of the power generation units in the current scheduling cycle, calculates the output change amplitude per unit time within the interval, and normalizes it as the source-side output adjustment range. At the same time, it collects the load response rate during the equipment operation period, obtains the power change rate data sequence of the load-side equipment, and obtains the source-load output change sequence. After calling the stable power generation resource screening set, the output parameter extraction submodule extracts the active power output time series of all power generation units within the current scheduling cycle. Assuming the scheduling cycle is 15 minutes, sampling is performed once per minute to obtain 15 continuous data points, and the minimum value is calculated. With the maximum value The difference between the two is the output range amplitude for that cycle, defined as: ; This amplitude value is compared with the maximum permissible variation capacity of all adjustable units within the dispatch area. Normalization process: ; For example, G1 unit within this cycle , And the maximum allowable adjustment step size within this region is ,but: ; Simultaneously, the power change curve of the equipment on the response side is retrieved from the load side, and the power slope between adjacent sampling points is extracted within this period. Let's assume a certain load equipment starts from the first... arrive The power is from Increase to The rate of change is: ; A load power change rate sequence is constructed using the difference values ​​of all time periods. The source-side output adjustment range is matched one by one with the load-side power change rate to finally form a source-load output change sequence, which provides input for downstream modules to construct slope difference data.

[0030] The slope construction submodule classifies and pairs the source-side output adjustment range with the load-side power change rate according to the source-load output change sequence, constructs source-load power time series respectively, extracts slope change points in each time period and converts them into a difference sequence under a unified time scale, constructs source-load power change slope difference sequence by period, and obtains source-load power slope difference value sequence. After receiving the source load output variation sequence, the variable slope construction submodule numbers each pair of source load combinations. Matching source-side rate of change With load-side change rate Plot the power time series graphs for each segment, and divide them into 15 time periods according to the scheduling cycle and a uniform sampling period (e.g., 1 minute). Calculate the differential slope within each time period. ,by The slope differences are constructed as a series, and the sum of the squares of all differences is obtained: ; This, in turn, is used to construct a coupling adaptation index together with the power generation response parameters. The calculation formula is as follows: ; in, This indicates the degree of deviation of the source load from the overall slope, reflecting the inconsistency in the macroscopic response trend. Represents the drasticness of micro-changes in the time domain; a larger value indicates data instability within a short period. (The denominator term contains...) Used to suppress the amplification effect of abnormal offsets in highly flexible resources. The negative impact of hysteresis resources on overall coupling stability is suppressed, and the three factors work together to form the coupling fit index.

[0031] Table 2. Calculation Parameters for Source-Load Coupled Adaptability Index

[0032] Substituting P1 into the calculation: The molecule is: ; The denominator is: ; ; This result indicates that the coupling fit index of the source load to P1 is: It belongs to the highly coordinated level, with the source and load showing high consistency in overall response rate and micro slope change, and has excellent response time and adjustment capability, making it a typical preferred choice.

[0033] The slope matching and filtering submodule calculates the average difference and fluctuation deviation rate within the scheduling cycle for each data set in the source-load power slope difference sequence, and constructs a coupling adaptation index by combining the response hysteresis time factor of the generation unit, using the formula: ; The coupling fit index value sequence of source-load pairs is obtained through calculation. Source-load pairs with index values ​​lower than the source-load coupling tolerance threshold are filtered out, a coupling effectiveness mark is established, and a source-load slope coupling list is obtained. in, This represents the source-load coupling fit index; the smaller the value, the higher the matching and coordination. This represents the normalized average of the source-side power change rate, and the data source is the ratio of the average value of the output difference sequence within the scheduling period to the theoretical maximum output rate. This represents the normalized average of the load-side power change rate, derived from the ratio of the power slope to the maximum load abrupt change rate over the corresponding time period. Indicates the first The difference in the rate of change of source load power at each time step, the source of which is and The absolute difference at that time step, This represents the number of time steps within the scheduling period, extracted by dividing the total scheduling period length by the data sampling period length. This represents the normalized average value indicating the flexibility of resource adjustment. It is derived from the mean of the difference between the maximum and minimum adjustment step sizes, normalized to the maximum allowable change value. The normalized value representing the source-side response hysteresis time is derived from the ratio of the response start delay to the standard response window length. The source-load coupling tolerance threshold is set based on the distribution density of the coupling adaptation index in successful matching cases of source-load pairs within historical scheduling cycles. Quantile analysis is used to fit the boundary of the dense index interval in the range of the 10th to 30th percentiles. The upper boundary of the interval is taken as the tolerance threshold allowed by the system. It is dynamically adjusted with different scheduling cycles. Since the coupling distribution is affected by the regional structure, the threshold setting does not use a fixed value, but is driven by the matching frequency to make differentiated settings.

[0034] After obtaining the coupling adaptation index sequence of all source-load pairs, the slope matching and filtering submodule reads the matching history database using the current scheduling cycle as the index. It retrieves the coupling index density distribution interval formed in the successfully coupled samples within the past 10 cycles and uses quantile analysis to take the maximum value between the 10th and 30th percentiles as the allowable source-load coupling tolerance threshold for the current cycle. For example, if the maximum value within this interval is Then the system settings are: ; The system assigns numbers to all source-load pairs. corresponding Value and Perform a comparison item by item, when Time markers are used to identify effective coupling pairs, and their source load numbers, response delay levels, priority values, and other labels are recorded to construct a source load slope coupling list. In this way, the micro-coordination matching requirements under complex resource dynamic change conditions can be mapped to an index evaluation mechanism, and dynamic threshold differentiation settings can be achieved, improving scheduling flexibility and actual executability.

[0035] Please see Figure 5 The electricity price range adjustment module includes: The delayed data extraction submodule calls the set of loads that have not been matched in the source-load slope coupling list, collects the unit power change of the load in the current scheduling cycle and the previous cycle, obtains the start time and arrival time of the output offset in the load response curve, calculates the adjustment range and response delay of each load device, establishes the correspondence between load number and adjustment-delay, and obtains the adjustment delay matching sequence. The delayed data extraction submodule performs data extraction operations on the load set marked "unmatched" in the source-load slope coupling list, performing differential analysis on the power sampling sequence of each load device in the current scheduling cycle and the previous scheduling cycle, with units of... Assuming 15 sampling points, a 15-dimensional sequence difference is formed. The change in unit power is extracted based on the difference between the point of maximum increase and the point of minimum power. Then, find the starting point of power change in the response curve of each device. With response point The time difference is defined as the response delay. For example, if load device L1's power starts to increase from the 3rd minute of the current cycle and stabilizes at the 7th minute, the response delay is... The total scheduling cycle is 15 minutes, which, after normalization, is as follows: ; Number each load device and assign its corresponding adjustment range. With normalization delay Binding is performed to generate an adjustment-delay mapping table, forming an adjustment delay matching sequence, which is used to call various variables in subsequent lag factor calculations.

[0036] The impact factor calculation submodule collects the real-time total load value and the predicted load value within the scheduling period based on the adjustment delay matching sequence, calculates the difference between the two as the load offset, and extracts the average response delay value, the maximum load adjustment amplitude, and the average change in load value within the current period using the following formula: ; Calculate and obtain the lag effect factor; in, Indicates the lagged effect factor. This represents the maximum load offset for all loads within the current period, derived from the maximum difference between the actual and predicted load values ​​among all unmatched loads. This represents the average load change within the scheduling period, derived from the arithmetic mean of the load fluctuation sequence within the period. This represents the normalized value of the average response delay for all unmatched loads within the current period. It is derived from the average of the differences between the response start and end times and is dimensionless after normalization to the maximum period length. This represents the normalized rate of change of load fluctuations, derived from the ratio of the standard deviation of load fluctuation amplitude per unit time to the maximum allowable rate of change of load. The load response density difference factor is derived from the offset ratio of the effective response load quantity to the total load equipment quantity within the scheduling cycle. The threshold for the electricity price change trigger ratio is set based on the critical point of system load sensitivity change within the electricity price stability range. The ratio of the 10-minute moving maximum load difference in the peak segment of intraday fluctuations in the system to the single-point electricity price response curvature is extracted and combined with the total number of responses within the cycle to form a regression distribution. The inflection point value is selected as the threshold and updated periodically. The impact factor calculation submodule collects the real-time total load value of the system during the period based on the adjusted delay matching sequence. Compared with model-predicted load values The difference between the two is the load offset: ; The maximum offset value among all unmatched loads is denoted as . For example, the maximum L2 offset of the load equipment is This value is used as the first term of the lag factor. Subsequently, the mean of the load fluctuation sequence of this equipment set is extracted. , set as And calculate the average response delay. and the standard deviation of unit load fluctuation Then use it with the maximum allowable rate of change. The ratio yields the rate of change of volatility: ; For example, the standard deviation of volatility is The maximum allowable rate of change ,have to Then, by statistically analyzing the difference between the number of effective response loads and the total number of devices within the scheduling cycle, the response density difference factor is obtained. This allows for the construction of a complete lagged impact factor: ; The formula consists of three parts, and the parameters in each part have a clear physical logic and numerical relationship: Part One This is used to measure the relative anomaly of the maximum actual load offset in the current cycle relative to the change in the system average load. The larger the value, the more the system prediction error is concentrated in a single or a few devices, and the stronger the instability. This item amplifies the interference weight of the offset device in square form, highlighting the impact of individual shocks on the overall prediction deviation of the system. Part Two Used to combine device average response delay With the rate of change of volatility The combined effect of these factors, if the response delay is longer, or the fluctuation during the response process is smaller (i.e., The smaller the value of the exponential function, the larger its value, further increasing the weight of the "lag potential" of this type of response. Conversely, if the response is slow but highly volatile, then due to... The increase leads to an expansion of the denominator and a reduction in the index weight, thereby reducing the sensitivity to the determination of overall lag risk. The addition of this item can better distinguish the systemic impact of the two types of delayed behavior: "slow and stable" and "slow and severe". Part Three The inverse moderating effect of the current response density on the degree of hysteresis is characterized by... If the value approaches 0, it indicates that the effective response load has covered most of the total equipment. If the value approaches 1, it indicates that the system has a strong collective response capability when the density is high, and the hysteresis can be absorbed by the collective characteristics. If the denominator is too large (insufficient density), the increase in the denominator will cause the term to tend to 0.8 or even lower, and the system's ability to resist hysteresis will be weakened.

[0037] The above three factors together constitute the final lagged impact factor. This indicator comprehensively evaluates three types of lag attributes: "local error concentration, response delay trend, and overall system response capability." The higher the value, the more likely the system has problems such as strong response lag, concentrated adjustment range, or insufficient response efficiency in the current cycle. Correspondingly, the sensitivity or flexibility of electricity price adjustment or resource regulation needs to be improved to prevent the lag effect from expanding and causing system imbalance.

[0038] Table 3. Calculation Parameters for Lag Effect Factor

[0039] Taking L2 as an example, substituting into the formula: The ratio of the squares of the first term is

[0040] The index term is

[0041] The second item is

[0042] Calculate the lag effect factor: ; This value indicates that the L2 device has a high degree of lag in the current period. Its high offset, high latency and volatility intensity together lead to a significant increase in the risk of system regulation and prediction offset. The value is used as the basis for the price boundary correction module to make judgments.

[0043] The price boundary correction submodule is based on the lag impact factor. It determines whether the lag impact factor is higher than the set threshold for the electricity price change trigger ratio. If the condition is met, it extracts the upper and lower limits of the price range in the previous period and the minimum and maximum response electricity value in the current period. It calculates the adjustment step size according to the floating ratio of the upper and lower limit boundaries and performs addition and subtraction operations on the upper and lower limits of the range in turn to obtain the dynamic adjustment parameters of the electricity price range. The price boundary correction submodule first sets the threshold for the ratio that triggers electricity price changes. This value originates from the ratio of load difference to electricity price slope in the daily peak load change stability experiment. It is derived from regression modeling using the number of effective responses across multiple periods, setting the current period as... The system sets the upper and lower limits of the price range for the previous period to... And to calculate the effective response electricity price range for the current period. If the calculated load is Then the ratio of the maximum to minimum price difference is used as the adjustment factor: ; The system adjusts the upper and lower boundaries of the floating range proportionally: ; After rounding to the nearest cent, the price range is adjusted as follows: This data is recorded as the dynamic adjustment parameters for the current period's electricity price range, used for updating matching rules in the price-sensitive dispatch window. Finally, a dynamic electricity price adjustment set is constructed for the integrated linkage module to load and prioritize updates.

[0044] Please see Figure 6 The integrated control and linkage module includes: The response parameter collection submodule calls the dynamic adjustment parameters of the electricity price range and the matched source load items in the source load slope coupling list, extracts the response delay and electricity price elasticity coefficient of each source load pair, classifies and marks them according to the scheduling time axis, and constructs the time-series response array of the source load pair in the current period to obtain the source load time-series response index set. The response parameter collection submodule calls the source-load combinations marked as "matched" in the list of dynamic adjustment parameters for electricity price ranges and source-load slope coupling, iterates through each source-load item, and reads the response delay of that combination. Electricity price elasticity coefficient Assuming a sampling period of 1 minute, each pair of combinations can construct a response array consisting of 15 time steps within a complete scheduling cycle. Each time step records whether the response triggering condition is met. Response delay value. This indicates the delay required from issuing the price adjustment signal to the actual commencement of load adjustment for that pair of combinations. For example, the response delay for group S1 is... If the group is unresponsive for the first two minutes of the dispatch cycle, it will be marked as responsive after the third minute. Electricity price elasticity coefficient. This indicates the sensitivity of the group to electricity price changes; the closer the value is to 1, the more effective the price adjustment signal is in regulating its behavior. A typical air conditioning load is a highly flexible load. Energy storage devices are mostly low-elasticity loads. Using the source-load pair number as the primary key, the system generates a structure array that records response delay, electricity price elasticity, and the marked status of each response point on the time axis, forming a source-load time-series response index set.

[0045] The adjustment window construction submodule extracts the response delay value and electricity price elasticity coefficient value based on the source-load time sequence response index set, calculates the intersection period with the current electricity price adjustment range, maps it to the time step in the time axis, determines whether each source-load pair falls into the intersection segment, marks it as adjustable, and generates a table of effective source-load adjustment segments. The adjustment window construction submodule extracts the parameters for each combination from the source load time-series response index set. and And match it with the upper and lower limits of the current cycle electricity price adjustment range (given by the upstream module), and set the adjustment range as... The response delay is incremented from the current time and intersected with the price adjustment window. If the starting time of the response delay of a certain source load group falls within this window, it is determined to be in an "adjustable" state, and is marked bit-by-bit on the time axis to generate a "Source Load Adjustment Effective Segment Table". Taking source load combination S3 as an example, the response delay... The electricity price range starts and ends at... The response trigger point is The adjustable range is left and right. The 7-minute time window; if the electricity price window is And the response delay is Then the source load responds to the initiation point Those still within the interval are marked as valid. After all combinations have been processed, the system generates a table of the effectiveness of source-load time axis adjustments, which is stored in the state matrix for subsequent priority calculations.

[0046] Table 4. Parameters of Source-Load Response Indicators and Electricity Price Adjustment Range

[0047] Taking S4 as an example, its response delay is Therefore, it is expected to have adjustment capabilities starting from the 3rd minute, within the range. An effective response is formed within the regulation range if the regulation range is defined as follows: If so, then all of S4 falls into it, which is an adjustable state.

[0048] The priority update submodule extracts the response delay and electricity price elasticity level values ​​based on the source load pairs marked as adjustable in the effective range table of source load regulation. According to the synchronicity and sensitivity level of the two in the regulation range, the priority index value is calculated. The adjustable source load items in the current cycle are reordered from high to low according to the index value to obtain the priority list of regulation execution. After determining the effectiveness of source load regulation, the priority update submodule aggregates and sorts all source load items in the "adjustable" state according to their response performance and electricity price perception capability. The system calculates the regulation priority index as follows. : ; In this formula, This indicates that the faster the regulatory response, the smaller the value; the inverse proportion is used as the weight in the denominator. This indicates that a higher response strength triggered by price changes is better, with a direct proportional effect placed in the numerator. Combining these two approaches ensures that source-load combinations with rapid and positive price responses are prioritized for scheduling. Taking S3 as an example, its... , The priority index is: ; Based on this calculation, all valid items are sorted to obtain: S3: ; S4: ; S1: ; S2: ; The system updates the source-load execution queue in the current cycle according to priority index from high to low, recording fields such as index, ranking, score, and source-load pair number, ultimately forming a control execution priority list, which is used for the final control output of the decision-making and scheduling window of the integrated control linkage module. This list has time-based positioning and dynamic responsiveness, supporting rapid iterative execution paths for different electricity price adjustment ranges.

[0049] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A virtual power plant generation-consumption-price collaborative optimization system based on an AI large-scale model, characterized in that, The system includes: The load confidence matching module, based on the target virtual power plant area, calls a hybrid model of convolutional neural network-long short-term memory network pre-trained based on load clustering labels and temperature control condition feature sequences. It performs boundary overlap rate calculation on the predicted load confidence interval and the power generation output response range to generate large model prediction mapping matching information. The load confidence matching module includes: The behavior data extraction submodule collects the active power time series, start-stop frequency and temperature control setpoint on the user side based on the target virtual power plant area, divides them into labels according to equipment category, and assigns them to cluster identifier sequences and temperature control feature sequences respectively. The cluster identifier and temperature control sequence are then bound to form a behavior sample vector set to generate a load behavior sample set. The load confidence prediction submodule, based on the load behavior sample set, calls the trained convolutional neural network-long short-term memory network model, combines the target prediction period, outputs the upper and lower confidence limits of each type of equipment within the period, establishes the confidence interval boundary sequence of the predicted load, records the corresponding time step of the prediction curve, and generates a confidence boundary value sequence. The boundary matching judgment submodule collects the standard deviation coefficient, cold start time and maximum ramp rate of the power generation unit based on the confidence boundary value sequence, constructs the upper and lower boundaries of the output response of each power generation unit, calculates the interval overlap rate between the predicted load confidence boundary and the output response boundary, and compares the overlap rate with the confidence overlap threshold to generate large model prediction mapping matching information. The resource stability mapping module extracts the output fluctuation rate, cold load recovery rate and equipment thermal inertia parameters of three consecutive control cycles from the set of power generation units with overlapping boundaries in the large model prediction mapping matching information, and filters the resource set that meets the stability change threshold, and generates a power generation resource stability screening set. The resource stabilization mapping module includes: The parameter extraction submodule marks the set of power generation units with overlapping boundaries in the large model prediction mapping matching information, collects the output records of each power generation unit in three consecutive control cycles, extracts the output data sequence, cold load response curve and thermal parameter change records during equipment operation, and calculates the output fluctuation rate, cold load recovery rate and equipment thermal inertia parameters respectively to obtain the set of periodic operation parameters. The fluctuation analysis submodule calls the output fluctuation rate, cold load recovery rate and equipment thermal inertia parameters in the set of periodic operation parameters, establishes a sliding window sequence according to time steps in three control cycles, calculates the sliding standard deviation corresponding to each parameter, takes the difference between the maximum and minimum values ​​of the standard deviation sequence as the fluctuation range, calculates the total resource fluctuation sequence, labels the resources according to the resource fluctuation sequence, and generates resource fluctuation label values. The resource screening submodule determines whether the volatility of each power generation unit is lower than the set resource stability change threshold based on the resource volatility label value, filters the resource units that meet the conditions, and records the control cycle number, stability identifier and volatility level to obtain the power generation resource stability screening set. The source-load capacity coupling module calls the output adjustment range of the power generation resource stability screening set, extracts the load change slope of the load-side equipment, compares the power slope of the source and the load, extracts the coupling terms that meet the slope difference tolerance threshold, and generates a source-load slope coupling list. The electricity price range adjustment module calls the set of loads that have not been matched in the source-load slope coupling list, calculates the lag influence factor, applies a step size adjustment to the upper and lower limits of the current electricity price range, and generates dynamic adjustment parameters for the electricity price range.

2. The AI-based large-scale virtual power plant generation-consumption-price collaborative optimization system according to claim 1, characterized in that, The large model prediction mapping matching information includes the distribution of predicted load confidence boundaries, output response tolerance range, and overlap rate compliance labels. The power generation resource stability screening set includes the regulation cycle stability score, historical thermal inertia variation amplitude, and output fluctuation threshold verification labels. The source-load slope coupling list includes a response regulation comparison table, slope matching level labels, and cycle matching validity records. The electricity price range dynamic adjustment parameters specifically include the step size adjustment coefficient, the upper and lower limit price variation amplitude, and the regulation cycle range identifier.

3. The AI-based large-scale virtual power plant generation-consumption-price collaborative optimization system according to claim 1, characterized in that, The source-load coupling module includes: The output parameter extraction submodule calls the set of power generation units in the stable screening set of power generation resources, extracts the minimum and maximum output values ​​of the power generation units in the current scheduling cycle, calculates the output change amplitude per unit time within the interval, and normalizes it as the source-side output adjustment range. At the same time, it collects the load response rate during the equipment operation period, obtains the power change rate data sequence of the load-side equipment, and obtains the source-load output change sequence. The slope construction submodule classifies and pairs the source-side output adjustment range with the load-side power change rate according to the source-load output change sequence, constructs source-load power time series respectively, extracts slope change points in each time period and converts them into a difference sequence under a unified time scale, constructs source-load power change slope difference sequence periodically, and obtains source-load power slope difference value sequence. The slope matching and filtering submodule calculates the average difference and fluctuation deviation rate within the scheduling cycle for each set of data in the source-load power slope difference sequence, and constructs a coupling adaptation index in combination with the response hysteresis time factor of the power generation unit. It then calculates and obtains the coupling adaptation index value sequence of the source-load pair, filters the source-load pairs whose index values ​​are lower than the source-load coupling tolerance threshold, establishes a coupling validity mark, and obtains the source-load slope coupling list.

4. The AI-based large-scale model-based virtual power plant generation-consumption-price collaborative optimization system according to claim 1, characterized in that, The electricity price range adjustment module includes: The delayed data extraction submodule calls the set of loads that have not been matched in the source-load slope coupling list, collects the unit power change of the load in the current scheduling cycle and the previous cycle, obtains the start time and arrival time of the output offset in the load response curve, calculates the adjustment range and response delay of each load device, establishes the correspondence between load number and adjustment-delay, and obtains the adjustment delay matching sequence. The impact factor calculation submodule collects the real-time total load value and the predicted load value within the scheduling period according to the adjustment delay matching sequence, calculates the difference between the two as the load offset, extracts the average response delay value of the current load set, the maximum load adjustment range and the average change of the load value within the current period, and calculates the lag impact factor. The price boundary correction submodule determines whether the lag factor is higher than the set electricity price change trigger ratio threshold based on the lag factor. If the condition is met, it extracts the upper and lower limits of the price range in the previous period and the minimum and maximum response electricity values ​​in the current period, calculates the adjustment step size according to the floating ratio of the upper and lower limit boundaries, and performs addition and subtraction operations on the upper and lower limits of the range in sequence to obtain the dynamic adjustment parameters of the electricity price range.

5. The AI-based large-scale virtual power plant generation-consumption-price collaborative optimization system according to claim 1, characterized in that, The system also includes: The integrated control and linkage module calls the response delay and price elasticity coefficient of the matching source load items in the power price range dynamic adjustment parameters and source load slope coupling list to construct the correspondence matrix between the control window and the price sensitive section. Based on whether it falls into the window execution section on the current time axis, it updates the priority execution order and generates the control execution priority list. The priority list for regulation execution includes source-load response ranking numbers, electricity price sensitivity weight values, and regulation window segment labels.

6. The virtual power plant generation-consumption-price collaborative optimization system based on an AI large model as described in claim 5, characterized in that, The integrated control and linkage module includes: The response parameter collection submodule calls the dynamically adjusted parameters of the electricity price range and the matched source load items in the source load slope coupling list, extracts the response delay and electricity price elasticity coefficient of each source load pair, classifies and marks them according to the scheduling time axis, and constructs the time-series response array of the source load pair in the current period to obtain the source load time-series response index set. The adjustment window construction submodule extracts the response delay value and the electricity price elasticity coefficient value based on the source-load time sequence response index set, calculates the intersection period with the current electricity price adjustment range, maps it to the time step in the time axis, determines whether each source-load pair falls into the intersection segment, marks it as adjustable, and generates a source-load adjustment effective segment table. The priority update submodule extracts the response delay and electricity price elasticity level values ​​based on the source-load pairs marked as adjustable in the effective range table of source-load regulation. According to the synchronicity and sensitivity level of the two in the regulation range, it calculates the priority index value and reorders the adjustable source-load items in the current cycle from high to low according to the index value to obtain the regulation execution priority list.

Images