Virtual power plant multi-element resource dynamic aggregation method considering source-load correlation

Abstract

The application discloses a virtual power plant multi-element resource dynamic aggregation method considering source-load correlation, and belongs to the technical field of power supply and distribution. The steps comprise the following steps: a source-side output feature and a load-side demand feature are extracted from a resource time sequence feature set, statistical correlation and time sequence correlation are calculated, and a source-load correlation feature is generated; a scene set reflecting a source-load joint change rule is constructed; time-effect layering and feasible region integration are performed on resource constraints; optimization and risk checking are performed in the aggregated feasible region to form a reliable dynamic aggregation instruction; and finally, correlation weight and resource boundary set are corrected by using execution feedback data. The method of the application can respond to power grid peak shaving demand with high reliability and economy by adopting the source-load correlation feature construction and joint change constraint generation coupling scene, and by fusing time-effect layering boundary to depict the aggregated feasible region.

Description

Technical Field

[0001] This invention belongs to the field of power supply and distribution technology, and in particular relates to a method for dynamic aggregation of multiple resources in a virtual power plant that considers source-load correlation. Background Technology

[0002] A virtual power plant is a virtual entity that aggregates distributed power sources, energy storage systems, and controllable loads scattered throughout the distribution network using advanced information and communication technologies and intelligent control methods. Externally, it appears as a dispatchable whole, participating in power ancillary services and the peak-shaving market. The multi-resource dynamic aggregation method involves jointly assessing the power and energy margins of these various resources over a specific time period to form an overall regulation capability that meets the grid's regulation needs, and generating control commands for individual resources. This technology is the core component of virtual power plant operation, directly affecting the accuracy and economy of peak-shaving response.

[0003] Existing research typically employs multi-resource aggregation methods based on deterministic resource models. These models treat the predicted output of each power generation unit and the baseline power of the load as known values, then superimpose equipment physical constraints and user contractual limits to form a fixed aggregated power range. Some methods introduce scenario analysis, establishing probability distribution or time series generation models for wind power, photovoltaics, and load respectively. Multiple independent scenarios are extracted to calculate the aggregated capacity, and then the aggregation boundary is obtained through quantile statistics. When dealing with source-load relationships, these methods often assume that the random variations on the generation and load sides are independent, or only implicitly consider a portion of the spontaneous interaction with the generation response within the total load.

[0004] The main shortcomings of the existing scheme are as follows: First, there is a correlation and time lag coupling between the fluctuation of power generation output and the change of adjustable load demand due to meteorological factors, production behavior or market driving. Ignoring this source-load correlation will cause the aggregation scenario to deviate excessively from the actual coupling mode, making the aggregation feasible domain estimation conservative or overly optimistic, thereby increasing the risk of peak shaving deviation. Second, the resource aggregation boundary is static or only updated at fixed cycles during operation, lacking a continuous learning and model correction mechanism based on actual execution feedback. As equipment performance degrades and user response habits change, the accuracy of the boundary gradually deteriorates, making it impossible to dynamically guarantee the fulfillment of aggregation commands. Summary of the Invention

[0005] Based on the shortcomings of the existing technology, the purpose of this invention is to provide a method for dynamic aggregation of multiple resources in a virtual power plant that considers source-load correlation. This method uses source-load correlation features to construct coupled scenarios and joint change constraints to generate them, and integrates time-dependent hierarchical boundaries to characterize the feasible domain for aggregation. This method can respond to the grid peak-shaving demand while taking into account both high reliability and economy.

[0006] The above objectives can be achieved through the following approach: A method for dynamic aggregation of multiple resources in a virtual power plant considering source-load correlation includes the following steps: The operation measurement information and control reachability characteristics of various resources within the virtual power plant are obtained, cleaned, and time-aligned to obtain a resource time-series feature set; Extract source-side output features and load-side demand features from the resource time-series feature set, calculate statistical correlation and time-series correlation, and generate source-load correlation features; Based on the source-load correlation characteristics, the resource time series feature set is mapped to a scenario to generate a source-load coupled scenario set. Resource constraint parsing is performed on each scenario in the source-load coupling scenario set to generate a resource constraint space set; The feasible region is integrated into the set of resource constraints to generate an aggregated feasible region. Obtain grid peak-shaving demand information and map it to the aggregated feasible domain for matching and optimization to generate dynamic aggregated solutions; Perform consistency and risk checks on the dynamic aggregation scheme to generate trusted dynamic aggregation instructions; The process involves issuing trusted dynamic aggregation instructions and obtaining execution feedback to form execution feedback data, and then updating the generation process of source-load correlation features and resource constraint space set based on the execution feedback data.

[0007] Preferably, the resource time-series feature set includes: The system acquires power measurement sequences, energy state sequences, and switch mode state sequences of various resources within a virtual power plant to generate operational measurement information. The system acquires communication latency information, controllable command type information, and response time window information of various resources within a virtual power plant to generate controllable reachability features. Missing data, anomalies, and time alignment are performed on operational measurement information and control reachable features to generate a resource time-series feature set.

[0008] Preferably, the source-load correlation features include: Identify the output change sequence corresponding to the power-generating resources from the resource time series feature set, and extract the fluctuation amplitude feature and the ramp feature to generate source-side output features; Identify demand change sequences corresponding to adjustable loads from resource time-series feature sets and extract transferable and reduceable quantity features to generate load-side demand features; Perform correlation calculations on the source-side output characteristics and load-side demand characteristics, and output the correlation coefficient matrix and hysteresis matrix; Based on preset correlation weights, the correlation coefficient matrix and the lag effect matrix are integrated to generate source-load correlation features.

[0009] Preferably, the source-load coupling scenario set includes: Construct joint variation constraints based on source-load correlation characteristics, and generate joint constraint parameters; Within a preset historical window of the resource temporal feature set, joint sampling is performed according to the joint constraint parameters to generate a source-load joint trajectory set; The feasibility of the source-load joint trajectory set is screened and the scene weights are output to generate the source-load coupled scene set.

[0010] Preferably, the generated resource constraint space set includes: Obtain the equipment rating information, operating status information, and user-side constraint information of each resource in the multi-source system to form a constraint input set; The constraint input set is mapped to power boundary, energy boundary and ramp boundary to form a resource boundary set; By binding the resource boundary set with the source-load coupling scenario set, a resource constraint space set is generated.

[0011] Preferably, the feasible domain for generating aggregation includes: The resource constraint space set is divided into layers according to a preset response time window to generate a time-sensitive layered constraint set; Cross-resource trajectory overlay and conflict resolution are performed on the time-sensitive hierarchical constraint set to generate an aggregated constraint trajectory set; Project the aggregated constraint trajectory set as an aggregated power boundary and an aggregated persistence boundary to form a boundary pair; Generate aggregate feasible regions using boundary pairs.

[0012] Preferably, the dynamic aggregation generation scheme includes: Obtain grid peak-shaving demand information and extract the demand for regulation direction, regulation magnitude and regulation duration to generate a peak-shaving demand vector; Input the peak-shaving demand vector into the aggregated feasible region to perform a feasibility determination, and output the feasible section; Within the feasible range, optimization is performed with the goal of limiting the preset peak-shaving deviation value and the preset resource switching cost, and a dynamic aggregation scheme is generated.

[0013] Preferably, the generation of trusted dynamic aggregation instructions includes: Substitute the dynamic aggregation scheme back into the source-load coupling scenario set for scenario consistency verification and generate consistency indicators. Substitute the dynamic aggregation scheme back into the resource constraint space set to verify constraint satisfaction and generate risk indicators. Based on consistency and risk indicators, realizability is calculated to generate available margin. Based on available margin, generate reliable dynamic aggregation instructions.

[0014] Preferably, the process of generating the source load correlation characteristics and resource constraint space set based on execution feedback data includes: Issue trusted dynamic aggregation commands and collect actual resource output sequences, actual response time sequences, and actual energy state sequences to generate execution feedback data; The execution feedback data is aligned with the resource time series feature set to identify deviations and the deviation patterns are extracted to generate a deviation feature set. Relevance weights are adjusted based on the deviation feature set; The resource boundary set is corrected based on the deviation feature set.

[0015] The present invention has the following advantages: This invention constructs source-load correlation features and joint change constraints, which can take into account the actual collaborative change patterns between power generation resources and load resources, thereby generating a high-fidelity source-load coupling scenario set and solving the problem of scenario distortion caused by neglecting the random correlation of source and load in traditional methods.

[0016] This invention performs hierarchical processing of resource aggregation constraints based on response time windows, generating a multi-dimensional aggregated feasible domain that includes fast response, constant response, and slow response layers. This enables the virtual power plant to characterize its adjustability at different time scales, significantly improving the dynamism and accuracy of resource potential assessment.

[0017] This invention ensures that the dynamic aggregation scheme has a high fulfillment rate in uncertain environments by back-substituting the aggregation scheme into the scenario for consistency verification and risk check, and iteratively correcting the instructions based on the availability margin. This enhances the support stability of the virtual power plant for grid peak shaving needs. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating a method for dynamic aggregation of multiple resources in a virtual power plant that considers source-load correlation. Figure 2 This is a schematic diagram of the source-load hysteresis cross-correlation curve in Example 1; Figure 3 This is a schematic diagram of a module in a virtual power plant dynamic aggregation system for multiple resources that considers source-load correlation. Detailed Implementation

[0019] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0020] Example 1: As Figure 1 As shown, the method for dynamic aggregation of multiple resources in a virtual power plant considering source-load correlation includes the following steps: S1. Obtain the operation measurement information and control reachability characteristics of multiple resources in the virtual power plant, and clean and align them with time to obtain the resource time series feature set; S2. Extract the source-side output characteristics and load-side demand characteristics from the resource time-series characteristic set, calculate the statistical correlation and time-series correlation, and generate source-load correlation characteristics; S3. Based on the source-load correlation characteristics, perform scenario-based mapping on the resource time-series feature set to generate a source-load coupled scenario set; S4. Perform resource constraint parsing on each scenario in the source-load coupling scenario set to generate a resource constraint space set; S5. Integrate the feasible regions of the resource constraint space set to generate an aggregated feasible region; S6. Obtain grid peak-shaving demand information and map the grid peak-shaving demand information to the aggregated feasible domain for matching and optimization, and generate dynamic aggregated solutions; S7. Perform consistency and risk checks on the dynamic aggregation scheme to generate trusted dynamic aggregation instructions; S8. Issue trusted dynamic aggregation instructions and obtain execution feedback to form execution feedback data. Update the generation process of source load correlation characteristics and resource constraint space set based on execution feedback data.

[0021] Step S1 includes the following steps: The system acquires power measurement sequences, energy state sequences, and switch mode state sequences of various resources within a virtual power plant to generate operational measurement information. The system acquires communication latency information, controllable command type information, and response time window information of various resources within a virtual power plant to generate controllable reachability features. Missing data, anomalies, and time alignment are performed on operational measurement information and control reachable features to generate a resource time-series feature set.

[0022] Operational measurement information includes power measurement sequences, energy status sequences, and switching mode status sequences. The power measurement sequence refers to the active power sampled values ​​of each resource at consecutive timestamps, denoted as... The unit is kilowatts, where i is the resource number. Let k be the sampling time. The energy state sequence refers to the time-varying record of remaining available or adjustable energy from resources such as energy storage and adjustable loads, denoted as […]. The unit is kilowatt-hour. The switch-mode state sequence refers to the discrete encoding of resource operation modes, such as shutdown, standby, and grid-connected charging, denoted as... It uses integer encoding.

[0023] Controllable reachability features include communication latency information, controllable command type information, and response time window information. Communication latency information represents the average round-trip delay and jitter range from when the aggregated control platform issues a command to when the resource side receives and begins execution, denoted as... The unit is milliseconds. Controllable instruction type information represents the set of control instructions that a resource can respond to, such as power regulation, mode switching, and direct switching, and is represented by a binary encoded vector. Description. The response time window refers to the time interval required for the output power to change from its current value to 90% of the target value after the resource begins executing the instruction. It is denoted as... The unit is milliseconds.

[0024] Missing data repair for power measurement sequences For gaps caused by communication interruptions or sensor malfunctions, linear interpolation of the preceding and following effective values ​​is used for filling. If the gap length exceeds 6 sampling intervals, random filling is performed based on the historical mean-variance prior of the same time period for the same type of day. Anomaly removal utilizes a sliding window to statistically analyze local means and standard deviations. Sample points deviating from the local mean by more than 3 times the standard deviation are marked as anomalies and replaced using linear interpolation between adjacent points. Time alignment unifies all time series from all resources onto a common time axis, using a combination of sample-and-hold and linear interpolation to resample to a fixed step size. Typically, 15 minutes are taken. The aligned operational measurement information and control reachability features are merged into a resource temporal feature set, denoted as... , where N is the total number of diverse resources.

[0025] Step S2 includes the following steps: Identify the output change sequence corresponding to the power-generating resources from the resource time series feature set, and extract the fluctuation amplitude feature and the ramp feature to generate source-side output features; Identify demand change sequences corresponding to adjustable loads from resource time-series feature sets and extract transferable and reduceable quantity features to generate load-side demand features; Perform correlation calculations on the source-side output characteristics and load-side demand characteristics, and output the correlation coefficient matrix and hysteresis matrix; Based on preset correlation weights, the correlation coefficient matrix and the lag effect matrix are integrated to generate source-load correlation features.

[0026] Powerable resources include photovoltaic power generation units, wind power generation units, etc., and their output change sequence is the power measurement sequence. For the i-th power-generating resource, the power standard deviation is calculated as a fluctuation amplitude characteristic within a time window of length W. The unit is kilowatt, where W is the length of the time window. The average power of the i-th power-generating resource within the time window W; the ramp-up characteristic is taken as the maximum absolute value of the rate of change of power between adjacent time points. Unit: kilowatts per minute For the i-th power-generating resource at the next sampling time The power measurement sequence. The source-side output characteristics are denoted as... .

[0027] Adjustable loads refer to industrial loads, commercial air conditioning clusters, etc., that have demand response capabilities; their demand change sequence is a power measurement sequence. Transferable quantity characteristics This represents the maximum adjustable power capacity of the load within a given time period, while maintaining total electricity consumption constant. It is determined based on the fluctuation envelope of historical demand sequences and user contract capacity, and is expressed in kilowatts. (Characteristics of adjustable capacity) This represents the maximum power that can be reduced without significantly impacting production or comfort, conservatively estimated using the 5th percentile of historical reduction events, expressed in kilowatts. Load-side demand characteristics are denoted as... .

[0028] Let the total number of power generation resources be The total number of load resources is Construct the correlation coefficient matrix Its elements The fluctuation amplitude in the i-th power generation resource feature vector Transferable quantity in the j-th load resource characteristic The Pearson correlation coefficient between them, i.e. Calculated from historical multi-day sample pairs, among which The covariance calculation function, The standard deviation of the volatility characteristics in the historical sample. This represents the standard deviation of transferable characteristics in the historical sample. (Lapsion effect matrix) elements For a pair of values ,in This represents the maximum absolute value of the cross-correlation coefficient between the power generation sequence and the load power sequence under different time delays. To obtain the lag step number corresponding to this maximum value, the step number increases in the direction that power generation leads load, and the lag time is... The calculation uses a cross-correlation function to iterate through the lag steps from... arrive , This is the maximum number of observation lag steps specified in advance.

[0029] Relevance weights include direct correlation weights. and lagged related weights ,satisfy For the combination of the i-th power generation resource and the j-th load resource, the source-load correlation characteristic element... Calculate using the following formula: ; Introduction The factor causes the coupling strength with a large lag to naturally decrease. The value ranges from 0 to 1 and is dimensionless. Source-load correlation characteristics. for matrix.

[0030] Step S3 includes the following steps: Construct joint variation constraints based on source-load correlation characteristics, and generate joint constraint parameters; Within a preset historical window of the resource temporal feature set, joint sampling is performed according to the joint constraint parameters to generate a source-load joint trajectory set; The feasibility of the source-load joint trajectory set is screened and the scene weights are output to generate the source-load coupled scene set.

[0031] Joint variation constraints describe the coordinated variation law of power generation resources and load resources in terms of power change. For all power generation resources and load resources, their power change vectors are considered. ,in, This represents the power change of the i-th power generation resource at time t, indicated by the superscript. Represents power generation resources, ; This represents the power change of the j-th load resource at time t, indicated by the superscript. Represents power generation resources, The increment is defined as the difference between adjacent time points. , , They represent Time and The power measurement sequence at time points. Joint constraint parameters include the mean vector. With covariance matrix Each element of the mean vector is the average of its respective historical power increment.

[0032] The generation-load cross-block elements in the covariance matrix are set according to the source-load correlation characteristics: the covariance between the i-th generation resource and the j-th load resource is... ,in and These represent the standard deviations (in kilowatts) and covariances (in square kilowatts) of the historical samples of power increments for the i-th power generation resource and the j-th load resource, respectively. The covariances between power generation resources and between load resources are directly derived from historical sample statistics. This is how the following is generated. Joint normal distribution As a joint change constraint.

[0033] A preset history window refers to a representative continuous time interval, the length of which is... Cover the entire daily cycle or multi-day pattern, such as the past 30 days. Select the start time from the historical time period. The actual power value at that moment is used as the initial state, and then joint distribution sampling is used to generate subsequent values. The power increment sequence of each step is accumulated to obtain the trajectory. : ; in, and They represent the s-th trajectory in Time and The source-load joint power state vector at time t; Indicates the s-th trajectory in The power increment vector obtained by sampling according to the joint normal distribution at each time step; the superscript s is the trajectory number, and the value is... Repeated sampling generates S independent trajectories, each containing the power time series of all generation and load resources. Forming a source-load joint trajectory set .

[0034] The feasibility screening process checks the following constraints on each trajectory: the output of power generation resources does not exceed their rated power and is not lower than the minimum technical output; the load power does not exceed the maximum available capacity and does not supply power in reverse; the state of charge of energy storage is maintained within a safe range; and the power change of each resource in adjacent time periods does not exceed the ramp limit. Trajectories that pass the screening are retained, and the number of retained trajectories is denoted as K. A scenario weight is assigned to each retained trajectory. Let be the probability density value of the trajectory under a multivariate normal distribution. Then, normalize the probability densities of all retained trajectories by summation to obtain the result that satisfies... And all terms are non-negative. Final source-load coupling scenario set. It consists of the preserved trajectory and the corresponding scene weights, i.e. The trajectories correspond one-to-one with the scenes, that is, the s-th trajectory corresponds to the s-th scene.

[0035] For example, a virtual power plant includes a photovoltaic power station with an installed capacity of 500kW, a wind farm with an installed capacity of 800kW, an energy storage system with a rated power of 200kW and a capacity of 400kWh, and an adjustable industrial load with a maximum adjustable power of 300kW. Operational measurement information is collected from 0:00 to 24:00 on a certain day, with time aligned to a specific step size. Within minutes, a resource time-series feature set is generated. Power generation resources are identified as photovoltaic (PV) and wind power, and load resources are identified as adjustable industrial loads. The PV fluctuation range is calculated within a 4-hour window. kW, climbing characteristics kW / 15min; Wind power fluctuation amplitude kW, climbing characteristics kW / 15min; Adjustable load transfer capacity kW, amount that can be reduced kW. The correlation coefficient matrix was calculated using a 30-day historical data. , .

[0036] like Figure 2 As shown, the lag effect matrix calculation yields a maximum cross-correlation value of 0.45 between photovoltaic power and load, corresponding to a lag of... The maximum cross-correlation between wind power and load is 0.20, corresponding to a lag. Step 1. Set relevance weights. Maximum lag steps In the source-load correlation characteristics, the photovoltaic-load term Wind power - load item In constructing the joint constraint parameters, the standard deviation of photovoltaic power increment is 25kW, the standard deviation of wind power increment is 50kW, and the standard deviation of load power increment is 35kW, thus obtaining the covariance between photovoltaic and load. kW², Covariance between wind power and load kW². Within a preset 30-day historical window, 1000 power trajectories are generated by sampling according to a joint normal distribution. After filtering based on resource power upper and lower limits, energy storage SOC safety range, and ramp-up constraints, 327 feasible trajectories are retained. The probability density is then calculated and normalized to obtain the weight of each scenario, forming a source-load coupling scenario set.

[0037] Step S4 includes the following steps: Obtain the equipment rating information, operating status information, and user-side constraint information of each resource in the multi-source system to form a constraint input set; The constraint input set is mapped to power boundary, energy boundary and ramp boundary to form a resource boundary set; By binding the resource boundary set with the source-load coupling scenario set, a resource constraint space set is generated.

[0038] The constraint input set is formed by acquiring the equipment rating information, operating status information, and user-side constraint information of each resource in the multi-source resource set. Equipment rating information refers to the limit operating parameters specified on the nameplate or design documents of the resource. For the i-th resource, the equipment rating information includes rated power. The unit is kilowatt, rated energy capacity. The unit is kilowatt-hour, applicable to energy storage resources, and the maximum continuous operating time. and minimum downtime All units are minutes.

[0039] Operational status information refers to the operational limits of a resource at the current moment or in a future timeframe, including current energy storage. The unit is kilowatt-hours, and the current operating mode is... It uses integer encoding and a health status conversion factor. Dimensionless, ranging from 0 to 1, calculated based on the equipment's cumulative runtime and maintenance records, and set according to the performance degradation curve provided by the equipment manufacturer. User-side constraint information refers to the restrictions proposed by the resource owner or operator for participating in aggregation regulation, including the lower limit of the power allowed for regulation. With power limit All units are kilowatts, adjustable lower energy limit. and upper limit The units are all kilowatt-hours, along with the upper limit of the permissible rate of change of power. The unit is kilowatts per minute, and this user-side constraint information is extracted from the terms of the bilateral response contract.

[0040] The constraint input set is mapped to power boundaries, energy boundaries, and ramp boundaries, forming a resource boundary set. For each resource i, the power boundary is a time series function, giving the allowable power output range at any time t. Based on the equipment's rated information, operating status information, and user-side constraint information, the lower limit of power is taken as the largest of the three sets of values: The upper bound of the power is taken as the minimum of the three values: In the formula This is a short-term overload factor, taken as 0.05, dimensionless. This factor is set based on the short-term overload capacity data provided by the equipment manufacturer.

[0041] The energy boundary, for energy storage and adjustable loads with energy storage characteristics, is expressed as the allowable range of stored energy at any time t. Lower energy level Where 0.1 is the deep discharge protection factor, the upper limit of energy. All figures are in kilowatt-hours. The ramp boundary is defined as the power change rate limit between adjacent control steps; the upper ramp boundary... Downhill boundary Take the same value as when climbing uphill, where The inherent ramp rate coefficient is expressed as the ratio of kilowatts per minute to rated power, and is determined by the type of equipment; for example, it is 0.5 minutes for gas turbines. -1 Energy storage converter takes 2.0 min. -1 The resource boundary set is denoted as... .

[0042] By binding the resource boundary set with the source-load coupled scenario set, a resource constraint space set is generated. As generated in step S3, each scene trajectory provides the power time series of all power generation and load resources. For resource i in scene s, its power trajectory value at time t is... Combined with the power boundary of the resource boundary set, a scene-specific power tolerance correction is generated: when Falling Outside of this, tighten the power boundary to... The intersection of the neighborhood centered at the slope constraint and the original boundary.

[0043] The energy boundary is also verified time-by-time based on the evolution of the energy storage state of charge in the scene trajectory, and the energy storage evolution follows the formula. ,in The time step is in hours, and the power is positive when discharging. and Let i and s represent the i-th resource in the s-th scenario, at time t and t, respectively. The energy state at any given moment. If If the lower or upper energy bound is reached, the discharge or chargeable power after that moment is forcibly set to zero until the energy state returns to the safe range. The resource constraint space set consists of the allowable power-energy-climbing joint constraints of all resources in all scenarios. Let the constraint space of resource i in scenario s be denoted as . In the formula, the power and energy boundaries after scene binding correction are superimposed with the scene index s; These represent the power variable, energy variable, and rate of climb variable within the constraint space, respectively. These represent the lower and upper bounds of the actual available power of the i-th resource in scenario s, after scenario binding correction. The lower and upper bounds of the actual available power of the i-th resource in scenario s, after scenario binding correction. This indicates the amount of power change within adjacent control steps.

[0044] Step S5 includes the following steps: The resource constraint space set is divided into layers according to a preset response time window to generate a time-sensitive layered constraint set; Cross-resource trajectory overlay and conflict resolution are performed on the time-sensitive hierarchical constraint set to generate an aggregated constraint trajectory set; Project the aggregated constraint trajectory set as an aggregated power boundary and an aggregated persistence boundary to form a boundary pair; Generate aggregate feasible regions using boundary pairs.

[0045] The resource constraint space set is layered according to a preset response time window to generate a time-stratified constraint set. The response time window is derived from the response time interval of each resource contained in the resource time-series feature set in step S1. The unit is milliseconds, but during the constraint process, the millisecond dimension needs to be converted to the index layer corresponding to the decision cycle of the control system. The decision step size of the control system is then set. Set the response time window for all resources to 15 minutes. The values ​​are mapped to Z time-sensitivity levels from smallest to largest, and the mapping method is as follows: Then resource i is assigned to the z-th layer, where Let be the response time threshold for the z-th layer, taken as 30 seconds, 5 minutes, and 15 minutes respectively, corresponding to the fast response layer, normal response layer, and slow response layer. The time-sensitivity layer constraint set is formed by merging the constraint spaces of each resource within each time-sensitivity layer under the source-load coupling scenario. The constraint set for the z-th time-sensitivity layer is denoted as . The same resource may have different available power adjustment ranges in different time-sensitivity layers because the fast response layer only considers... Resources that respond within seconds are typically energy storage and heat pumps, while the slow response layer encompasses all resources.

[0046] Cross-resource trajectory overlay and conflict resolution are performed on the time-sensitive hierarchical constraint set to generate an aggregated constraint trajectory set. (This is repeated in the original text.) In this context, for a given scenario s and time t, the upper bound of the aggregated power at that time is obtained by summing the upper bounds of the power allowable for all resources within that layer. The lower bound of the aggregation power is obtained by summing the lower bounds of the power allowances. In terms of energy, the upper and lower bounds of the allowable energy state of energy storage resources are summed to obtain the upper bound of aggregate energy. and lower bound of aggregated energy All units are kilowatt-hours.

[0047] Conflict resolution occurs when response conflicts arise between resources within the same time horizon. A typical conflict manifests as the ramp-up capability of some resources being limited by the abrupt change in the scene trajectory, resulting in time spikes in the available power margins for both upscaling and downscaling. Conflict resolution addresses this by introducing flexible coupling relaxation variables. The upper and lower bounds of the polymerization power were corrected to and ,in satisfy , A resolution scaling factor of 0.05 is used; it is dimensionless and its value is based on the 5% statistical threshold of the average amplitude of power spikes in multi-scenario simulations. The aggregated constraint trajectory set consists of the aggregated power and aggregated energy sequences of each time-dependent layer in each scenario after conflict resolution, denoted as... .

[0048] The aggregated constraint trajectory set is projected as an aggregated power boundary and an aggregated persistence boundary, forming a boundary pair. The aggregated power boundary is directly extracted from the power dimension envelope of the aggregated constraint trajectory set. For the time-dependent layer z, in the aggregated power trajectories of multiple scenarios s=1, 2, ..., K, the scenario-weighted quantile is taken as the final boundary: the upper bound of the aggregated power. Lower bound of polymerization power In the formula Indicates scene weight The weighted q quantiles, with q values ​​of 0.95 and 0.05 representing conservative and tight upper and lower bounds respectively, are in kilowatts.

[0049] The polymerization duration boundary characterizes the length of time that polymerization power can be sustained at a given power level. It is derived from the polymerization energy boundary: for any power adjustment P, the upper limit of polymerization duration... Take the satisfaction The largest value, For the duration to be solved, aggregate the lower bound of duration. Take the satisfaction The largest The value, both are in minutes, of which For time integration variables, , For each time level z and time u, generate the upper and lower bounds of the aggregated power. For each time level z, generate a boundary pair. .

[0050] Boundary pairs are used to generate the aggregated feasible region. The aggregated feasible region unifies the boundary pairs of each time-dependent layer into a closed region on the power-duration two-dimensional plane. For the fast response layer z=1, the constant response layer z=2, and the slow response layer z=3, the power boundaries of each layer are first plotted with the power adjustment amount as the horizontal axis and the duration as the vertical axis. and The corresponding adjustable power range, and then the polymerization persistence boundary. and Curves are plotted to represent the upper and lower limits of the duration for each power point within this range. The zero point on the horizontal axis corresponds to zero power adjustment; a positive direction indicates an upward adjustment, and a negative direction indicates a downward adjustment. The feasible region is the union of all time-limited layer boundaries, i.e., for each... Coordinates, if there exists a layer z such that and If a point does not exceed the upper bound of the aggregation corresponding to that layer, then the point belongs to the aggregation feasible region, denoted as . The mathematical expression for the aggregate feasible region is: ; Where Z represents the total number of aging layers, and P represents the power regulation amount, in kilowatts. Duration, in minutes. and Let z be the lower and upper bounds of the aggregation power of the z-th time layer. This represents the upper bound of the aggregation persistence of the z-th time-dependent layer at power P. The persistence boundary in the downward adjustment direction is also considered. When matching peak-shaving requirements, the absolute value is incorporated into this expression.

[0051] For example, based on the virtual power plant in the aforementioned step example, it includes 500kW photovoltaic power, 800kW wind power, 200kW / 400kWh energy storage, and 300kW adjustable industrial load. In the equipment rating information, the photovoltaic rated power of 500kW has no energy capacity, the wind power rated power of 800kW has no energy capacity, the energy storage rated power of 200kW has a rated energy capacity of 400kWh, the maximum continuous operating time is 120 minutes, the minimum downtime is 0 minutes, and the adjustable load rated power is the maximum adjustable capacity of 300kW. In the operating status information, the current energy of the energy storage is 300kWh, and the health status conversion factor is... .

[0052] In the user-side constraint information, the allowable adjustable power range for energy storage is 0 to 200 kW, the adjustable energy range is 40 kWh to 400 kWh, and the ramping limit is 200 kW / min; the allowable adjustable load has a lower limit of 50 kW, an upper limit of 300 kW, and a ramping limit of 60 kW / min. After mapping the constraint input set, the resource boundaries, such as the lower bound of energy storage power, are... Upper limit of power The resource boundary set is bound to the source-load coupling scenario set (327 scenarios generated by S3). The energy storage power boundary is not lowered in the high-output photovoltaic scenario, but the energy storage absorbs the excess power. The ramp constraint limits the charge-discharge switching rate.

[0053] Based on response time windows, the energy storage response time range of 30-100 milliseconds is classified into the fast response layer (within the 30-second threshold); the adjustable load response time range of 2-10 seconds is classified into the fast response layer; the wind power pitch response of 5-15 minutes is classified into the normal response layer (within the 5-minute threshold); and photovoltaic power without fast control measures is classified into the slow response layer (within the 15-minute threshold). For the fast response layer, cross-resource trajectory overlays are performed. For a given scenario at a given time, the upper limit of energy storage power is 200kW, the upper limit of load downscaling capability is 300kW, and the upper limit of aggregated power is 500kW. After conflict resolution, the values ​​are taken as... The upper limit of the modified polymerization power was obtained as 475kW.

[0054] When projecting the aggregated constraint trajectory set, the 95th quantile is used to form the aggregated power boundary. Continuity boundary calculations show that a 400kWh energy storage capacity can be sustained for 120 minutes at full power (200kW). With the addition of load shedding capacity, the fast-speed layer's up-regulation duration reaches 240 minutes at 100kW. The final aggregated feasible region gives a range where the fast-speed layer can provide up-regulation power from 0 to 475kW for 10 seconds to 240 minutes; the constant-speed layer provides up-regulation from 0 to 400kW for 5 minutes or more from wind power; and the slow-speed layer provides up-regulation from 0 to 100kW for 15 minutes or more from photovoltaic load shedding. The union of these three layers forms the complete aggregated feasible region. .

[0055] Step S6 includes the following steps: Obtain grid peak-shaving demand information and extract the demand for regulation direction, regulation magnitude and regulation duration to generate a peak-shaving demand vector; Input the peak-shaving demand vector into the aggregated feasible region to perform a feasibility determination, and output the feasible section; Within the feasible range, optimization is performed with the goal of limiting the preset peak-shaving deviation value and the preset resource switching cost, and a dynamic aggregation scheme is generated.

[0056] The process involves acquiring grid peak-shaving demand information and extracting the direction, magnitude, and duration of regulation demand to generate a peak-shaving demand vector. Grid peak-shaving demand information is a standardized data message transmitted from the grid dispatch center to the virtual power plant aggregation control platform. This message includes a demand timing description: the start time of the request, the end time of the request, the target power regulation value, and the direction of regulation. The direction of regulation demand indicates whether the virtual power plant needs to perform an upward or downward regulation operation, denoted by the symbol δ. δ=+1 represents an upward regulation, meaning the virtual power plant needs to increase grid-connected power or decrease power consumption; δ=-1 represents a downward regulation, meaning the virtual power plant needs to decrease grid-connected power or increase power consumption. This symbol is dimensionless. The magnitude of regulation demand is the expected power change of the grid, denoted as... The unit is kilowatt. Take a positive value. The duration for which the power change remains uninterrupted is denoted as (the value is missing from the original text). The unit is minutes. Extract the above three items from the message and arrange them in the order of adjustment direction sign, amplitude, and duration to form a peak-shaving demand vector. ,in , , .

[0057] Input the peak-shaving demand vector into the aggregated feasible region, perform a feasibility assessment, and output the feasible segment. Aggregated feasible region Generated by step S2, it is a closed region on the power regulation-duration plane, mathematically expressed as: The horizontal axis P represents the power adjustment amount, with a positive direction corresponding to an upward adjustment and a negative direction corresponding to a downward adjustment, in kilowatts; the vertical axis... This represents the duration, in minutes. and These are the lower and upper bounds of the aggregation power at the z-th aging layer, respectively. This represents the upper bound of the polymerization persistence of the z-th aging layer at power P.

[0058] Map the peak-shaving demand vector to a target point in the aggregate feasible region coordinate system. Its x-coordinate y-axis Feasibility assessment involves constructing a relaxed neighborhood around the target point, with the power half-width of the neighborhood taken as... Duration half-width ,in This is the preset power deviation tolerance coefficient. This is a preset tolerance coefficient for duration deviation; both are dimensionless. Based on the allowable deviation rate of 0.05 for exemption from assessment in the power grid ancillary services market, The 95th percentile of the historical peak-shaving command execution time deviation is set to 0.10. This neighborhood and the aggregated feasible region... The intersection of these points constitutes the feasible section. , represented as: ; If the intersection is not empty, the feasibility check passes, and the feasible segment is output. Used for subsequent optimization; if the intersection is empty, successively... and At the same time, increase the step size by 0.01 until the intersection is not empty, and output the new intersection as the feasible segment.

[0059] Within the feasible range, an optimization strategy is generated based on a preset peak-shaving deviation value and a preset resource switching cost constraint. The preset peak-shaving deviation value includes a power deviation cost coefficient. Duration deviation cost coefficient The unit is yuan, and the pricing is based on the tiered pricing of penalties for grid peak-shaving deviation assessments. Relative deviation per unit squared Yuan per unit squared relative deviation. For any point within the feasible section. Peak shaving deviation cost Calculated by weighted summation of the square of the relative power deviation and the square of the relative duration deviation: ; In the formula The absolute value of the power adjustment is expressed in kilowatts. Duration, in minutes. Preset resource switching costs include unit energy cost for each timeframe. The unit is yuan per kilowatt-hour, and the fixed switching cost for activating a certain time-layer is also included. The unit is yuan. The unit energy cost for different time-sensitivity layers is set based on the typical marginal adjustment cost of this type of resource: fast response layer Yuan per kilowatt-hour, constant speed response layer Yuan per kilowatt-hour, slow response layer Yuan per kilowatt-hour; fixed switching cost The cost is set at 5 yuan per instance, representing the amortization of lifespan loss and communication overhead caused by changes in resource status.

[0060] During optimization, 0-1 integer variables are introduced. Indicates whether the z-th time-limit layer is invoked; a continuous variable. This represents the amount of power regulation undertaken by the z-th time layer. The duration provided by the z-th time layer must satisfy power balance. ,and At the same time, the power and duration of each layer must fall within that layer. Within the contribution boundary, i.e. It should be between the upper and lower power bounds of the z-th layer. No more than this layer The upper bound of the lower bound. The total regulation energy of a certain layer. Unit is kilowatt-hour, resource switching cost Calculate using the following formula: ; Overall Optimization Objective Function The optimization variable is , , and the synthesized , constraint is Furthermore, the power and duration components satisfy the relationship formed by the aggregation boundaries of each time-limited layer. This optimization is a mixed-integer nonlinear programming problem, solved using a branch-and-bound algorithm combined with a sequential quadratic programming algorithm. The result is the decision variable values ​​corresponding to the minimum objective function, including whether each time-limited layer is activated and the power allocation of each layer. and duration The dynamic aggregation scheme is generated from these decision variables, taking the form of a set of power adjustment instructions for each participating resource: if a certain time-sensitive layer is activated, the resources within the layer undergo power decomposition based on their respective capacity proportions and dynamic response characteristics to obtain the adjustment target value for each resource. The corresponding execution period extends from the moment the demand begins. For minutes, resources in the non-enabled layer will continue to operate as planned. The dynamic aggregation scheme is denoted as... ,in The unit is kilowatt.

[0061] For example, based on the aggregated feasible domain formed by the aforementioned steps, the virtual power plant includes 500kW of photovoltaic power, 800kW of wind power, 200kW / 400kWh of energy storage, and 300kW of adjustable load. The fast layer includes energy storage and adjustable load, the constant-speed layer includes wind power, and the slow-speed layer includes photovoltaic load shedding. At 17:55 on a certain day, the virtual power plant control platform receives peak-shaving demand information from the power grid: a request to provide up-regulation services between 18:00 and 19:00, with a regulation range of 300kW, maintained for 60 minutes. A peak-shaving demand vector is generated. Mapping target point .

[0062] Within the aggregated feasible domain, the fast on-layer regulation capability (0-475kW) can be sustained for 10 seconds to 240 minutes, the constant-speed on-layer regulation capability (0-400kW) can be sustained for 5 minutes or more, and the slow-speed on-layer regulation capability (0-100kW) can be sustained for 15 minutes or more. , Constructing feasible sections as kilowatts and Minutes area and The intersection of the two layers. Since the target point (300, 60) falls within the boundaries of both the fast layer and the normal layer, the feasible section is not empty and includes multiple possibilities such as being undertaken by a pure fast layer, being undertaken by a pure normal layer, or being a mixture of both layers.

[0063] Optimize within the feasible range, and set... Yuan, Yuan, Yuan / kWh Yuan / kWh Yuan / kWh Yuan. Evaluation of pure constant-rate layer schemes: , kW, min, kWh, , Yuan, total target 41 yuan; pure fast layer solution: , kW, min, kWh, , The total target is 95 yuan. The optimization algorithm iterates through the hybrid schemes and finds that the constant-speed layer alone is the globally optimal solution. Therefore, a dynamic aggregation scheme is generated: the wind farm increases its power by 300kW from 18:00 to 19:00, energy storage and adjustable loads remain unchanged, and photovoltaics operate at maximum power point tracking. This scheme... Formal output, including wind farm power commands kW, for 60 minutes This indicates the baseline power of the wind farm under the original planned operating mode; there are no incremental instructions for other resources.

[0064] Step S7 includes the following steps: Substitute the dynamic aggregation scheme back into the source-load coupling scenario set for scenario consistency verification and generate consistency indicators. Substitute the dynamic aggregation scheme back into the resource constraint space set to verify constraint satisfaction and generate risk indicators. Based on consistency and risk indicators, realizability is calculated to generate available margin. Based on available margin, generate reliable dynamic aggregation instructions.

[0065] The dynamic aggregation scheme is then substituted back into the source-load coupling scenario set for scenario consistency verification, generating consistency metrics. Generated by step S6, wherein Let be the power command value for resource i during time period t, in kilowatts. The command covers a time period that extends from the start of peak demand to the optimized duration. (Source-load coupling scenario set) Includes K joint power trajectories and their corresponding scenario weights Each trajectory provides the time-series power values ​​of all power generation resources and load resources within a preset historical window. Scenario consistency verification requires evaluating the degree to which the actual adjustment effect of the aggregated solution matches the peak-shaving needs of the power grid in each scenario.

[0066] For scenario s, define the actual execution power of resource i. The instruction value is the result after being constrained by the concentrated power boundary of the resource-constrained space, i.e. ,in The function represents when When take a, when If the condition is met, choose b; otherwise, choose x. and These represent the lower and upper bounds of the power of resource i in scenario s in step S2, respectively, both in kilowatts. The peak-shaving demand vector is defined with the output of generated power to the grid as the positive direction. Medium to high adjustment Lower the corresponding Adjustment range requirements Unit: kilowatt; regulating continuous demand. Unit: minutes.

[0067] The actual net regulation power generated by the aggregation scheme in scenario s is The formula involves summing all power generation and load resources to obtain the change in total grid-connected power, expressed in kilowatts. Based on the demand direction of the adjustment range, the active power adjustment in the same direction is calculated. This value should ideally be close to .

[0068] Introducing the power deviation tolerance factor defined in step S6 and duration deviation tolerance coefficient When scenario s satisfies the existence of a certain continuous time interval Completely included within the peak demand period, interval length And at every moment within this interval, the following condition is met. At that time, identify the consistency of the scenario. Set to 1 if true, otherwise set to 0. Consistency index We obtain the weighted summation based on the weights of all scenarios: ; This indicator is dimensionless and ranges from 0 to 1. It represents the combined probability of the aggregation scheme meeting peak-shaving requirements in terms of both time and magnitude within a set of scenarios.

[0069] The dynamic aggregation scheme is then substituted back into the resource constraint space set for constraint satisfaction verification, generating risk indicators. The resource constraint space set is generated in step S4, representing the constraint space of resource i in scenario s. The safe operating range of the resource in terms of power, energy, and ramp rate was defined. Constraint satisfaction verification examined whether the execution trajectory of the scheme violated the hard boundaries in this space.

[0070] For scenario s, define a power violation indicator variable. exist or The value is 1 if the hill climb is violated, and 0 otherwise; define a hill climb violation indicator variable. When the absolute value of the power command difference between adjacent control steps exceeds the ramp boundary. The value is 1 if the condition is met, and 0 otherwise; define an energy violation indicator variable. For resources with energy state constraints, when executing instructions causes a change in the simulated energy state... Beyond the energy boundary The value is 1 if the condition is met, and 0 otherwise. The time step is converted to hours. Comprehensive risk identification for scenario s. The value is taken as the maximum of the violation indications of all resources at all check times. If any resource has a flag of 1 at any time, then... ,on the contrary Risk indicators Calculated from scene weights: ; This dimensionless metric represents the probability that an aggregation scheme will trigger at least one violation of a resource constraint in a set of scenarios.

[0071] Based on consistency and risk indicators, realizability is calculated to generate availability margin. Realizability represents the overall probability that the aggregation scheme can effectively meet the grid peak-shaving needs without violating any resource constraints in a multi-scenario stochastic environment. Defined as the sum of scenario weights that simultaneously satisfy the conditions of scenario consistency qualification and have no constrained risk occurrence, i.e.: ; In the formula and These are the scenario consistency identifier and risk identifier obtained from the aforementioned steps, respectively. As scene weight, and This indicator is dimensionless, and its calculated value is directly used as the availability margin. The availability margin reflects the reliability of the solution's fulfillment under uncertain conditions; the closer the value is to 1, the more stable and reliable the solution is.

[0072] Generate trusted dynamic aggregation instructions based on availability margin. Set a trust threshold. This threshold is determined based on the historical pass rate requirements for virtual power plant aggregation participation in the power grid peak-shaving market and combined with engineering simulation experience. When the feasibility... At that time, the dynamic aggregation scheme generated in step S3 will be used. Directly mark it as a trusted dynamic aggregation instruction and add an execution priority label before outputting. When At that time, conservative adjustment is initiated: the adjustment range demand in the peak-shaving demand vector is adjusted to... At the same time, maintain the need for adjustment direction. and regulating continuous demand If unchanged, re-enter step S3 to generate a revised dynamic aggregation scheme, and perform the consistency and risk check of step S4 on the revised scheme again. Repeat this iteration until the reliability of the revised scheme reaches a certain level. The process terminates when the adjusted amplitude is lower than 0.1 times the lower limit of the original amplitude; the final solution that satisfies the usable margin condition or reaches the lower limit obtained in the last iteration is taken as the reliable dynamic aggregation instruction.

[0073] The format of the trusted dynamic aggregation command is consistent with the dynamic aggregation scheme, which includes the power command sequence of each resource, but additionally includes the available margin value and iteration round record for reference when the virtual power plant is executed.

[0074] For example, continuing the dynamic aggregation scheme obtained from the previous steps, the wind farm increases its power by 300kW for 60 minutes from 18:00 to 19:00, while other resources do not participate in this adjustment. The source-load coupling scenario set obtains a total of 327 scenarios from step S3, with each scenario having a weight. Substituting this scheme back into each scenario, we calculate the actual wind power output: For scenario s, the wind power baseline... Instruction value Upper limit of wind power The wind power generated in scenario s is the actual power generated. Aggregation adjustment amount .by Corresponding to 285~315kW, Corresponding to checks lasting 54-66 minutes, a total of 281 scenarios achieved a continuous power level above 285kW for at least 54 minutes, resulting in... .

[0075] During constraint satisfaction verification, it was checked whether the instructions exceeded the resource power boundary. It was found that in 46 scenarios, insufficient wind speed caused... This is considered a power violation risk; all other resources are not in violation. Corresponding to these 46 scenarios, When calculating realizability, consideration is given to cases that simultaneously satisfy consistency and have no risk. and The scenario yielded 240 results. The realizability is below the credibility threshold of 0.85, thus triggering a conservative adjustment, reducing the adjustment range to [amount missing]. kW, re-enter step S5 to optimize and obtain a new scheme, the wind farm is increased by 259kW, and the feasibility is checked again and the feasibility is increased to 0.88, reaching the threshold. At this time, the reliable dynamic aggregation command is output: the wind farm increases the power by 259kW from 18:00 to 19:00, and the available margin of the tag is 0.88.

[0076] Step S8 includes the following steps: Issue trusted dynamic aggregation commands and collect actual resource output sequences, actual response time sequences, and actual energy state sequences to generate execution feedback data; The execution feedback data is aligned with the resource time series feature set to identify deviations and the deviation patterns are extracted to generate a deviation feature set. Relevance weights are adjusted based on the deviation feature set; The resource boundary set is corrected based on the deviation feature set.

[0077] A trusted dynamic aggregation command is issued, and the actual output sequence, actual response time sequence, and actual energy state sequence of the resources are collected to generate execution feedback data. The trusted dynamic aggregation command is generated in step S7 and contains the power command sequence of each participating resource. The unit is kilowatt, and the available margin value is also specified. The instructions are issued through the communication link between the virtual power plant control platform and the local resource controller. The communication protocol adopts the IEC61850 manufacturing message specification and is mapped to the TCP / IP network.

[0078] The actual power output sequence of a resource is a time series of active power collected from the local resource measurement device and adjusted according to instructions, denoted as... Unit: kilowatt, sampling time Align with the resource timing feature set, maintaining a 15-minute interval. The actual response time series records the time elapsed for each resource from the arrival of the instruction until its output power change reaches 90% of the instruction's target change, denoted as... The unit is seconds, and the measurement is based on the time difference between the local controller trigger signal and the power change detection signal. The actual energy state sequence, for resources with energy storage or energy reserves, records the remaining energy value collected during command execution, denoted as... The unit is kilowatt-hour. Execution feedback data consists of combinations of the above three types of sequences, stored as a data table indexed by resource number and timestamp. .

[0079] The execution feedback data is aligned with the resource time-series feature set to identify deviations, and deviation patterns are extracted to generate a deviation feature set. Generated in step S1, containing power measurement sequences from the same historical period. Energy state sequence and communication latency information Etc. Deviation alignment uses each timestamp in the execution feedback data as an anchor point, retrieves historical sample sets belonging to the same date label and sampling time with consistent operating mode identifiers in the resource time series feature set, and calculates the power arithmetic mean of this sample set. With the mean of energy states If there are fewer than 5 historical samples, the 10 most recent identical time points will be used to supplement them.

[0080] Power deviation The unit is kilowatts, reflecting the degree to which the actual output deviates from the expected operating point when no instruction is given after the instruction is executed. Response time deviation. The unit is seconds, where This represents the average communication delay for the resource within the resource time-series feature set. Energy deviation. The unit is kilowatt-hour. Deviation patterns are extracted by statistically analyzing the mean and standard deviation of these deviations: for each resource i, the mean power deviation is calculated. and standard deviation ; mean response time deviation and standard deviation Mean energy deviation and standard deviation ,in The number of valid sampling points for this resource in the execution feedback data.

[0081] In addition, the changes in the coupling strength between power generation resources and load resources during actual implementation are extracted, i.e., the actual power increment sequence is calculated. The Pearson correlation coefficients between them form the actual correlation coefficient matrix. Peak matrix of hysteresis cross-correlation and the correlation coefficient matrix in step S2 and the hysteresis matrix By comparison, the correlation deviation matrix is ​​obtained. and Deviation feature set It consists of all these statistics, including as well as and The elements.

[0082] The correlation weights are adjusted based on the deviation feature set. The correlation weights are preset coefficients used in step S2 to fuse the correlation coefficient matrix and the lag effect matrix. and ,satisfy The deviation characteristics reflect whether the lag correlation between actual power generation and load has strengthened or weakened compared to historical records, as shown by the correlation deviation matrix. The weights are adjusted based on the trend of the largest lag correlation value. The average change in the lag effect is defined. ,in yes The element represents the change in the maximum cross-correlation coefficient between power generation resource i and load resource j, and is dimensionless.

[0083] like If the value exceeds the preset change threshold of 0.08, it indicates a significant change in the hysteresis correlation structure. In this case, corrections will be made according to the following rules: Set as ,in for The average of the absolute values ​​of all elements in the set. for The average of the absolute values ​​of all elements in the set. This is a sign function, outputting +1 or -1; the formula ensures... The change should not exceed 0.3 times the original value. Then... Updated to and to and A truncation protection mechanism is implemented to ensure that both weights fall within the interval [0.15, 0.85]. This interval is set based on empirical statistics from multiple rounds of peak-shaving events to prevent a single abnormal event from causing a weight to approach zero. If it does not exceed 0.08, then and It remains unchanged.

[0084] Resource boundary set correction based on deviation feature set. Generated by step S4, including power boundaries Energy boundary and the boundary of the slope The correction process utilizes the power deviation statistic and response time deviation, centered on the deviation characteristics, to adaptively shrink the boundary. For the power boundary, the relative coefficient of power deviation is calculated. Dimensionless Rated power of the equipment, in kilowatts.

[0085] when This means that the actual output is below the baseline, indicating an additional shortfall in the upward adjustment capability. Therefore, the upper limit of the power output is revised to... ;when This means that the actual output is biased higher than the baseline, indicating that the downward adjustment capability is limited, and the lower limit of power is revised to... In both cases, if If the sign is independent of the expected adjustment direction, the original boundary remains unchanged. The ramp boundary is corrected based on the response time deviation: a response time degradation coefficient is defined. Dimensionless, updating the original climbing boundary to This formula is based on the first-order response assumption, meaning that the effective ramp rate decreases proportionally as the response slows down. The energy boundary is corrected based on the relative magnitude of the mean energy deviation: if resource i is an energy storage type and Negative deviation exceeding rated energy capacity If the actual energy is 0.05 times lower than the historical average for the same period, then the upper bound of the energy level will be lowered to [value missing]. The lower energy boundary shifts upward to Both equations use kilowatt-hours as the unit, maintaining consistent dimensions. The corrected resource boundary set replaces the original boundary set and is used for the re-run of subsequent steps S1 to S4, forming a closed-loop self-updating mechanism.

[0086] For example, based on the trusted dynamic aggregation command generated in step S7, the wind farm increases its power by 259kW from 18:00 to 19:00, while the remaining resources remain in their original state. After the command is issued, the virtual power plant control platform collects execution feedback data, obtaining 4 sets of sampling points at 15-minute intervals. The actual output sequence of the wind farm is [245, 252, 260, 248]kW, and the actual response time sequence is [12.3, 11.8, 12.0, 11.5] seconds. No energy state sequence of the wind power is recorded.

[0087] The historical average daily wind power output for the same period in the resource time series feature set is [4, 8, 6, 7] kW. The calculated power deviation is [241, 244, 254, 241] kW. Since the wind power command target is 259 kW, the actual average adjustment deviation is -14 kW, i.e. kW, standard deviation kW. Mean response time deviation Assume historical communication delay seconds, then Seconds. The lag correlation between power generation and load in the correlation deviation matrix. Less than 0.08, correlation weight Maintain the original value.

[0088] Power boundary correction: Wind farm rated power 800kW, wind power boundary correction factor ,because Upper limit of power Originally 800kW, revised to kW. Climbing Boundary: Wind power climbing boundary correction factor The original climbing limit of 40kW / min has been revised to kW / min. The revised resource boundary set will be used in the next peak demand response process, enabling the upper limit of the aggregated feasible region's power and ramping capability to track the actual operating characteristics of the resources in real time.

[0089] Example 2: Based on the same inventive concept, such as Figure 3 As shown, the present invention also provides a virtual power plant multi-resource dynamic aggregation system that considers source-load correlation, including: The data acquisition module is used to acquire the operation measurement information and control reachability characteristics of various resources in the virtual power plant, and to clean and time-align them to obtain the resource time-series feature set. The correlation modeling module is used to extract source-side output features and load-side demand features from the resource time-series feature set, calculate statistical correlation and time-series correlation, and generate source-load correlation features. The scene generation module is used to perform scene-based mapping of resource temporal feature sets based on source-load correlation characteristics, and generate source-load coupled scene sets. The constraint resolution module is used to perform resource constraint resolution on each scenario in the source-load coupling scenario set and generate a resource constraint space set. The feasible region integration module is used to integrate the feasible regions of the resource constraint space set to generate an aggregated feasible region. The aggregation optimization module is used to obtain grid peak-shaving demand information and map the grid peak-shaving demand information to the aggregation feasible domain for matching and optimization, generating dynamic aggregation solutions; The verification output module is used to perform consistency verification and risk verification on the dynamic aggregation scheme and generate a trusted dynamic aggregation instruction. The online update module is used to issue trusted dynamic aggregation instructions and obtain execution feedback to form execution feedback data. Based on the execution feedback data, the module updates the generation process of source load correlation characteristics and resource constraint space set.

Claims

1. A method for dynamic aggregation of multiple resources in a virtual power plant considering source-load correlation, characterized by the following steps: include: The operation measurement information and control reachability characteristics of various resources within the virtual power plant are obtained, cleaned, and time-aligned to obtain a resource time-series feature set; Extract source-side output features and load-side demand features from the resource time-series feature set, calculate statistical correlations and time-series correlations, and generate source-load correlation features, including: Identify the output change sequence corresponding to the power-generating resources from the resource time series feature set, and extract the fluctuation amplitude feature and the ramp feature to generate source-side output features; Identify demand change sequences corresponding to adjustable loads from resource time-series feature sets and extract transferable and reduceable quantity features to generate load-side demand features; Perform correlation calculations on the source-side output characteristics and load-side demand characteristics, and output the correlation coefficient matrix and hysteresis matrix; Based on the preset correlation weights, the correlation coefficient matrix and the lag effect matrix are fused to generate source load correlation features; Based on the source-load correlation characteristics, the resource time series feature set is mapped to a scenario to generate a source-load coupled scenario set. Resource constraint parsing is performed on each scenario in the source-load coupling scenario set to generate a resource constraint space set; The feasible region is integrated into the set of resource constraints to generate an aggregated feasible region. Obtain grid peak-shaving demand information and map it to the aggregated feasible domain for matching and optimization to generate dynamic aggregated solutions; Perform consistency and risk checks on the dynamic aggregation scheme to generate trusted dynamic aggregation instructions; The process involves issuing trusted dynamic aggregation instructions and obtaining execution feedback to form execution feedback data, and then updating the generation process of source-load correlation features and resource constraint space set based on the execution feedback data.

2. The method for dynamic aggregation of multiple resources in a virtual power plant considering source-load correlation according to claim 1, characterized in that, The obtained resource time series feature set includes: The system acquires power measurement sequences, energy state sequences, and switch mode state sequences of various resources within a virtual power plant to generate operational measurement information. The system acquires communication latency information, controllable command type information, and response time window information of various resources within a virtual power plant to generate controllable reachability features. Missing data, anomalies, and time alignment are performed on operational measurement information and control reachable features to generate a resource time-series feature set.

3. The method for dynamic aggregation of multiple resources in a virtual power plant considering source-load correlation according to claim 1, characterized in that, The generated source-load coupling scene set includes: Construct joint variation constraints based on source-load correlation characteristics, and generate joint constraint parameters; Within a preset historical window of the resource temporal feature set, joint sampling is performed according to the joint constraint parameters to generate a source-load joint trajectory set; The feasibility of the source-load joint trajectory set is screened and the scene weights are output to generate the source-load coupled scene set.

4. The method for dynamic aggregation of multiple resources in a virtual power plant considering source-load correlation according to claim 1, characterized in that, The generated resource constraint space set includes: Obtain the equipment rating information, operating status information, and user-side constraint information of each resource in the multi-source system to form a constraint input set; The constraint input set is mapped to power boundary, energy boundary and ramp boundary to form a resource boundary set; By binding the resource boundary set with the source-load coupling scenario set, a resource constraint space set is generated.

5. The method for dynamic aggregation of multiple resources in a virtual power plant considering source-load correlation according to claim 1, characterized in that, The generated aggregate feasible domain includes: The resource constraint space set is divided into layers according to a preset response time window to generate a time-sensitive layered constraint set; Cross-resource trajectory overlay and conflict resolution are performed on the time-sensitive hierarchical constraint set to generate an aggregated constraint trajectory set; Project the aggregated constraint trajectory set as an aggregated power boundary and an aggregated persistence boundary to form a boundary pair; Generate aggregate feasible regions using boundary pairs.

6. The method for dynamic aggregation of multiple resources in a virtual power plant considering source-load correlation according to claim 1, characterized in that, The generation of dynamic aggregation schemes includes: Obtain grid peak-shaving demand information and extract the demand for regulation direction, regulation magnitude and regulation duration to generate a peak-shaving demand vector; Input the peak-shaving demand vector into the aggregated feasible region to perform a feasibility determination, and output the feasible section; Within the feasible range, optimization is performed with the goal of limiting the preset peak-shaving deviation value and the preset resource switching cost, and a dynamic aggregation scheme is generated.

7. The method for dynamic aggregation of multiple resources in a virtual power plant considering source-load correlation according to claim 1, characterized in that, The generation of trusted dynamic aggregation instructions includes: Substitute the dynamic aggregation scheme back into the source-load coupling scenario set for scenario consistency verification and generate consistency indicators. Substitute the dynamic aggregation scheme back into the resource constraint space set to verify constraint satisfaction and generate risk indicators. Based on consistency and risk indicators, realizability is calculated to generate available margin. Based on available margin, generate reliable dynamic aggregation instructions.

8. The method for dynamic aggregation of multiple resources in a virtual power plant considering source-load correlation according to claim 7, characterized in that, The process of generating the source load correlation features and resource constraint space set based on execution feedback data includes: Issue trusted dynamic aggregation commands and collect actual resource output sequences, actual response time sequences, and actual energy state sequences to generate execution feedback data; The execution feedback data is aligned with the resource time series feature set to identify deviations and the deviation patterns are extracted to generate a deviation feature set. Relevance weights are adjusted based on the deviation feature set; The resource boundary set is corrected based on the deviation feature set.

Images