A source-load-storage collaborative optimization control method and system for a new energy subject

Abstract

The application discloses a source-load-storage collaborative optimization control method and system for a new energy main body, and relates to the technical field of power grid control.The method comprises the following steps: constructing an energy storage device aggregation feasible region and combining a preset adjustment demand vector to generate a minimum adjustment capacity constraint; constructing a multi-time scale collaborative optimization model comprising a day-ahead scheduling layer, an intra-day rolling layer and a real-time control layer, and optimizing and solving based on the capacity constraint; when solving, extracting the day-ahead reserve margin and the intra-day power feasible region of the day-ahead scheduling layer and the intra-day rolling layer as the boundary constraints of the intra-day rolling layer and the real-time control layer; calculating the deviation penalty cost of the intra-day rolling layer and the real-time control layer and superimposing the objective functions of the day-ahead scheduling layer and the intra-day rolling layer to iteratively solve, and issuing the collaborative control instruction after convergence.The application is used to solve the problem that the prior art cannot quantize the boundary limited degree, resulting in the problem that the reserve margin cannot be adaptively adjusted under extreme fluctuations and the collaborative control capacity is limited.

Description

Technical Field

[0001] This invention relates to the field of power grid control technology, and more specifically, to a source-load-storage coordinated optimization control method and system for new energy entities. Background Technology

[0002] Existing source-load-storage coordinated optimization scheduling schemes mostly adopt a multi-timescale control architecture of day-ahead, intraday, and real-time to cope with the uncertain fluctuations in demand at both the source and load ends under a high proportion of new energy access.

[0003] For example, the invention application with announcement number CN117595401A discloses a source-grid-load-storage coordination method, system, and terminal based on multi-scale rolling optimization. It performs preliminary control through day-ahead scheduling, intraday rolling, and real-time coordination, and performs secondary prediction and feedback correction of the intraday rolling power plan based on the power operation status of the real-time coordination process. At the same time, it corrects the day-ahead scheduling plan based on the intraday power situation to reduce the cumulative impact of prediction errors.

[0004] However, the aforementioned existing technologies still have some technical problems in practical applications: On the one hand, when coordinating information and correcting commands across time scales, the physical power deviation value at the surface level is often directly used as the superimposed correction amount, failing to map the changes in the underlying control boundary into the economic cost in the objective function. This makes it difficult for the upper-level scheduling plan to proactively perceive the limited risks of the lower-level feasible space during execution, thus making it difficult to adaptively expand the reserve margin and ensure the system's coordinated control capability under continuous extreme scenarios. On the other hand, in the face of extreme disturbance scenarios, due to the lack of dynamic verification of the multi-dimensional adjustment boundaries of the energy storage cluster, when the system control demand exceeds the physical limits of the equipment, it is easy to cause constraint conflicts, leading to deadlock in the model solution or overload of field equipment. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a source-load-storage coordinated optimization control method and system for new energy entities. By constructing an aggregated feasible domain to generate minimum adjustment capability constraints, and superimposing the deviation penalty cost under the boundary constraints onto the corresponding upper-level objective function, the present invention addresses the problems of deadlock in solving problems and difficulty in quantifying the degree of boundary constraints in multi-timescale coordination, which leads to the inability to adaptively adjust reserve margin and limited global coordinated control capabilities.

[0006] To achieve the above objectives, this invention provides a source-load-storage coordinated optimization control method for new energy entities. Through closed-loop coordination of aggregated feasible regions and cross-scale optimization models, it achieves globally robust scheduling under high-proportion new energy access: In the physical boundary definition stage, an aggregated feasible region is first constructed based on the operating status and multi-dimensional adjustment capabilities of energy storage devices. Then, the minimum adjustment capability constraints of each energy storage device are extracted by combining a preset adjustment demand vector, thereby completing the preliminary matching of adjustment demand and device physical response capability and establishing the physical executable boundary. In the collaborative solution stage, a multi-timescale collaborative optimization model is constructed, including a day-ahead scheduling layer, an intraday rolling layer, and a real-time control layer. A cross-scale iterative mechanism is introduced during the model solution process: the day-ahead reserve margin and intraday power feasible regions generated by the day-ahead scheduling layer and the intraday rolling layer are extracted and used as dynamic boundary constraints for the intraday rolling layer and the real-time control layer, respectively, ensuring the nested consistency of the decision space. Simultaneously, the deviation penalty costs of the intraday rolling layer and the real-time control layer under the corresponding boundary constraints are calculated and then inversely superimposed onto the objective functions of the day-ahead scheduling layer and the intraday rolling layer for iterative solution until convergence and the collaborative control command is issued.

[0007] This invention proactively transforms the risk of physical overruns in a short timescale into a penalty term in a long timescale objective function through reverse feedback of physical boundary verification and cross-scale economic penalty costs. This prompts the scheduling plan to actively reserve response margins during the optimization process, effectively avoiding the technical problem that existing scheduling plans are prone to deviating from the actual response capabilities of equipment, thus leading to operational overruns. This improves the physical executability of control commands and the robustness of system operation.

[0008] In a preferred embodiment, the components of the preset adjustment demand vector include power demand, demand duration, and allowable delay; the step of obtaining the adjustment demand vector includes: performing extreme value extrapolation fitting and probability density estimation on the tail and main body of the error distribution in the historical prediction error data of new energy, and combining them to obtain a proposed distribution for scenario sampling; generating an initial scenario set based on the proposed distribution and clustering it to obtain a typical scenario set; extracting the net load gap of the typical scenario set, analyzing the peak value and fluctuation characteristics of the net load gap, and mapping it to the adjustment demand vector.

[0009] In a preferred embodiment, the clustering to obtain a typical scenario set includes: calculating the importance weights of the initial scenario set under the error distribution and the proposal distribution; using the importance weights, correcting the occurrence probability of extreme scenarios in the initial scenario set, and clustering the corrected initial scenario set into the typical scenario set.

[0010] In a preferred embodiment, the step of obtaining the deviation penalty cost of the real-time control layer includes: calculating the power deviation of the real-time control layer's decision output based on the reference trajectory generated by the intraday rolling layer in the current period; calculating the sensitivity coefficient of the objective function of the real-time control layer relative to the intraday power feasible domain boundary; and using the product of the power deviation and the sensitivity coefficient as the deviation penalty cost of the real-time control layer.

[0011] In a preferred embodiment, the step of obtaining the deviation penalty cost of the intraday rolling layer includes: obtaining the intraday source load prediction power under a typical scenario set; calculating the power deviation between the prediction power and the planned output of the day-ahead scheduling layer in the current period as the required reserve capacity; calculating the reserve capacity deficit based on the required reserve capacity and the day-ahead reserve margin; calculating the proportion of the reserve capacity deficit to the initial range of the intraday power feasible domain; and applying a nonlinear weighting to the deviation penalty cost of the real-time control layer to obtain the deviation penalty cost of the intraday rolling layer.

[0012] In a preferred embodiment, the construction of the aggregated feasible domain based on the operating status and multi-dimensional adjustment capabilities of the energy storage devices includes: obtaining the initial state of charge of each energy storage device in the current cycle, and combining the charging and discharging power boundaries, continuous discharge time and dynamic response characteristics of each energy storage device to construct a device feasible domain including the dimensions of power, time and ramp rate; aggregating each of the device feasible domains to obtain the aggregated feasible domain.

[0013] In a preferred embodiment, the step of extracting the minimum regulation capability constraint of each energy storage device includes: comparing a preset regulation demand vector with an aggregated feasible region, extracting responsive regulation demands that do not exceed the aggregated feasible region; and mapping the responsive regulation demands to the corresponding target energy storage device based on the maximum charge / discharge power and discharge depth of each energy storage device, and using them as the minimum regulation capability constraint of the target energy storage device.

[0014] In a preferred embodiment, the selection step of the target energy storage device includes: extracting the component values ​​of the responsive adjustment demand in each dimension of the aggregated feasible domain, and comparing them with the corresponding preset judgment thresholds to obtain the over-limit dimensions; when there are multiple over-limit dimensions, determining the core over-limit dimension according to a preset priority order, and matching the corresponding target energy storage device according to the adjustment characteristics of the core over-limit dimension.

[0015] In a preferred embodiment, the objective function of the day-ahead scheduling layer includes the day-ahead integrated operating cost and the deviation penalty cost of the intraday rolling layer; the objective function of the intraday rolling layer includes the intraday scheduling operating cost, the day-ahead plan tracking deviation cost, and the deviation penalty cost of the real-time control layer; the objective function of the real-time control layer includes the real-time power tracking deviation cost and the energy storage regulation cost.

[0016] A source-load-storage coordinated optimization control system for new energy entities includes the following modules: a constraint generation module, used to construct an aggregated feasible region based on the operating status and multi-dimensional adjustment capabilities of energy storage devices, and extract the minimum adjustment capability constraints of each energy storage device in combination with a preset adjustment demand vector; a coordinated optimization module, used to construct a multi-timescale coordinated optimization model including a day-ahead scheduling layer, an intraday rolling layer, and a real-time control layer, and to perform optimization solutions based on the minimum adjustment capability constraints; during the solution process, the day-ahead reserve margin generated by the day-ahead scheduling layer is used as the boundary constraint of the intraday rolling layer, and the intraday power feasible region generated by the intraday rolling layer is used as the boundary constraint of the real-time control layer; the deviation penalty cost of the intraday rolling layer and the real-time control layer under the corresponding boundary constraints is calculated, and then inversely superimposed into the objective functions of the day-ahead scheduling layer and the intraday rolling layer, respectively, and iteratively solved until convergence, and then a coordinated control command is issued.

[0017] The technical effects and advantages of the source-load-storage coordinated optimization control method and system for new energy entities, as described in this invention, are as follows: 1. This invention constructs an aggregated feasible domain based on the multi-dimensional adjustment capabilities of energy storage devices, and generates minimum adjustment capability constraints by combining a preset adjustment demand vector. By introducing physical boundary verification before model solving, it effectively avoids the risk of solution deadlock or field equipment exceeding limits caused by exceeding the control demand limit, and ensures the physical executability and system operation safety of source-load-storage collaborative optimization control under extreme operating conditions.

[0018] 2. This invention calculates the deviation penalty cost between the intraday rolling layer and the real-time control layer under corresponding boundary constraints, and then iteratively solves the problem by adding these costs to the objective functions of the day-ahead scheduling layer and the intraday rolling layer, respectively. By constructing a cross-timescale feedback transmission mechanism driven by economic costs, the upper-level scheduling model can proactively perceive the degree of constraint of the lower-level feasible space and adaptively expand the reserve margin, thereby improving the global optimization capability and robustness of the new power grid multi-timescale scheduling architecture in response to uncertain fluctuations in source and load. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of a source-load-storage coordinated optimization control method for new energy entities, provided as an embodiment of the present invention.

[0020] Figure 2 A schematic diagram comparing the probability density fitting of historical prediction error data for new energy provided in an embodiment of the present invention.

[0021] Figure 3 This is a schematic diagram of the iterative convergence trajectory of the multi-timescale collaborative optimization model provided in an embodiment of the present invention.

[0022] Figure 4This is a schematic diagram of a source-load-storage coordinated optimization control system module for new energy entities, provided as an embodiment of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0024] Example 1, Figure 1 This invention presents a source-load-storage coordinated optimization control method for new energy entities, comprising the following steps: Step 1: Construct an aggregated feasible region based on the operating status and multi-dimensional adjustment capabilities of energy storage devices, and extract the minimum adjustment capability constraints of each energy storage device by combining the preset adjustment demand vector.

[0025] 1.1 Based on the operating status and multi-dimensional adjustment capabilities of energy storage devices, an aggregated feasible domain is constructed, including: 1.1.1 Obtain the initial state of charge of each energy storage device in the current cycle, and combine the charge / discharge power boundaries, continuous discharge time, and dynamic response characteristics of each energy storage device to construct a device feasible region including power, time, and ramp rate dimensions, as follows: The system bus of the power industry-specific telecommunications management system (i.e., TMS2.0 version) reads real-time parameters reported by the underlying battery management system (BMS) at a high frequency of milliseconds. The real-time parameters include the state of charge of the battery cells, the measured internal resistance of the BMS, and the real-time thermodynamic temperature of the cells.

[0026] Using the measured internal resistance of the BMS, the real-time thermodynamic temperature boundary of the battery cell, the upper and lower limits of the battery's safe voltage, and the extreme threshold of the safe state of charge as inputs, the functional state (SOF) estimation method is adopted. First, based on the first-order RC equivalent circuit and Ohm's law for closed circuits, the first limiting current allowed without touching the upper and lower limits of the battery's safe voltage is derived and calculated. Second, based on the ampere-hour integral principle, the second limiting current allowed to be discharged before reaching the extreme threshold of the safe state of charge is calculated. Third, according to the temperature and current safety derating mapping relationship specified by the battery manufacturer, the third limiting current limited by the thermal boundary at the current real-time thermodynamic temperature of the battery cell is extracted. Finally, the smallest absolute value among the three limiting currents is extracted as the final output of the algorithm, that is, the maximum allowable charge and discharge current of the underlying hardware. The maximum allowable charge and discharge current is multiplied by the current terminal voltage to obtain the maximum charge and discharge power extreme value, which is used as the charge and discharge power boundary.

[0027] Multiply the current state of charge (SOC) by the rated capacity of the energy storage station to calculate the current absolute available power. Divide the current absolute available power by the rated operating power to obtain the physical available time that physically supports continuous discharge to the minimum safe depth of discharge of the battery, which is taken as the maximum continuous discharge time. Obtain the nameplate parameters and factory test reports of the energy storage inverters (PCS) of each energy storage device, and take the step response delay time (in seconds) and maximum ramp rate (MW / min) of the hardware inverter as the dynamic response characteristics.

[0028] The day-ahead dispatchable charging and discharging power, continuous discharge time, and ramp rate of the energy storage system are used as independent coordinate axes in a three-dimensional decision space. Then, based on the step response delay time in the dynamic response characteristics, the first... The constraint coefficient matrix of each energy storage cell is used to define the normal vector direction and spatial translation of each boundary surface of the polyhedron, thereby characterizing the mutual exclusion relationship of the charging and discharging states of the device, the conversion relationship between multidimensional physical quantities, and the inherent response hysteresis effect of the underlying hardware; the mathematical expression of the constraint coefficient matrix is ​​as follows: (1) in, For the first The constraint coefficient matrix of each energy storage unit. The set time step for a single scheduling operation (in seconds). For the first The step response delay time of each energy storage cell.

[0029] Subsequently, the calculated maximum charge / discharge power extreme value, maximum continuous discharge time, and maximum ramp rate are used as the maximum extreme intercepts of the corresponding dimensions to construct a boundary column vector. The first four rows of the boundary column vector are composed of the maximum discharge power extreme value, the maximum charge power extreme value, the maximum continuous discharge time, and the maximum ramp rate, respectively, and the fifth row is set to a constant of 0. Considering that battery physical aging will reduce the operating boundary, when the measured health status (SOH value) is determined to decrease by real-time reading through the BMS, the percentage of the measured SOH value to the factory SOH value is directly used as the attenuation coefficient, which is multiplied by the maximum extreme intercept of each dimension to achieve synchronous reduction of the physical extreme value.

[0030] Finally, since the charging / discharging power boundary, maximum continuous discharge time, and maximum ramp rate are all physically linear bounded constraints, these constraints correspond to a set of closed linear half-spaces in geometric space. According to linear algebra and polyhedral computation theory, the intersection of a finite number of linear half-spaces in three-dimensional Euclidean space must form a bounded geometric topology with straight boundary lines and in the form of a discrete convex polyhedron. Therefore, by using the constraint coefficient matrix and boundary column vectors, and intersecting multiple linear inequalities representing the physical upper limits of each dimension as boundary surfaces, the three-dimensional feasible region of the device enveloping the safe operating range of the device can be obtained. The mathematical expression of the three-dimensional feasible region is as follows: (2) in, To characterize the first The real-time state column vector of each energy storage cell, whose internal components correspond to the charge / discharge power (MW), continuous discharge time (in min), and ramp rate (MW / min); subscript Characterizing the node designation of heterogeneous energy storage devices; For the first The boundary column vector of each energy storage cell.

[0031] 1.1.2 Aggregate the feasible regions of each device to obtain the aggregated feasible region, as follows: A Minkowski algorithm based on vertex traversal is employed to superimpose spatial geometric vectors onto the feasible regions of each heterogeneous energy storage unit. Specifically: since the three-dimensional convex polyhedron corresponding to the feasible region of each unit is essentially an H-representation of a polyhedron (half-space / inequality constraint), this embodiment uses vertex enumeration technology from computational geometry to transform it into a V-representation (set of vertex coordinates). Using the set of vertex coordinates corresponding to the feasible region of each energy storage unit as input to the algorithm, a vertex is extracted from each vertex coordinate set. Linear addition of vertices within all possible combinations of extracted vertices is performed in the same dimension. The convex hull of the new coordinate point set generated after all additions is extracted to obtain the superimposed geometric space boundary, i.e., the aggregated feasible region. The aggregated feasible region inherits the three-dimensional spatial attributes of each unit's feasible region in the physical dimension. The extreme values ​​of the projection of its outer convex hull onto each coordinate axis constitute the aggregated maximum charge / discharge power boundary, the aggregated maximum continuous discharge time boundary, and the aggregated maximum ramp rate boundary for day-ahead master station control.

[0032] In actual engineering deployment, the H-representation can be transformed into V-representation by calling a solver configured with a standard computational geometry function library (such as cddlib, lrslib, or similar geometric computation packages). By using the constraint coefficient matrix and boundary column vector as algorithm input, the underlying encapsulated dual description method or reverse search algorithm is used for automatic solution, directly extracting and outputting the coordinates of all real topological vertices of the polyhedron that satisfy the global constraints, thereby obtaining the set of vertex coordinates of the feasible domain of each device.

[0033] 1.2 Obtain a preset adjustment demand vector including power demand, demand duration, and allowable delay, including: 1.2.1 Extreme value extrapolation fitting and probability density estimation are performed on the tail and main body of the error distribution in the historical prediction error data of new energy, respectively, and combined to obtain a proposed distribution for scenario sampling, as follows: Historical prediction error data of new energy is obtained by regulating the historical database of the main station. The error data specifically includes the power deviation between the day-ahead predicted active power output curve and the actual operating output curve of wind power and photovoltaic units in the past year. The historical prediction error data is sorted in ascending order according to the absolute value. The 95th percentile of the sequence is set as the extreme value division threshold. Data with an absolute error value less than or equal to the extreme value division threshold is divided into the main part, and data with an absolute error value greater than the extreme value division threshold is divided into the tail part.

[0034] A probability density function describing the main body of the prediction error data is constructed using kernel density estimation (KDE). The formula for the probability density function of the main body is as follows: (3) in, Let be the probability density function of the main body. For prediction error, The main data sample size For Gaussian kernel function, These are historical error sample points; To smooth the bandwidth, it is specifically a value calculated using Scott's rule based on the total number of samples in the main part of the error data and the standard deviation of the error data in that part. Simultaneously, for the segmented tail data, the maximum likelihood estimation method is used to fit the optimal shape and scale parameters. Based on the shape and scale parameters, the probability density function of the tail data is constructed using the generalized Pareto distribution (GPD) to extrapolate the extreme values ​​of extreme error events with heavy tails in the error data. The probability density function formula for the tail data is as follows: (4) in, Let be the probability density function of the tail data. For shape parameters, This is a scale parameter (unit: MW).

[0035] Based on empirical weighting coefficients, the probability density functions of the main and tail portions of the prediction error data are combined to obtain the proposed distribution for scene sampling. The calculation formula is as follows: (5) in, For prediction error variables The probability density function of the proposal distribution; This is an empirical weighting parameter, with a value ranging from 0 to 1. It is recommended when the installed capacity of new energy sources such as wind and solar power accounts for more than 50% of the system's total capacity. The optimal value is 0.2 to 0.3, to enhance the weight of extreme tail distributions and prevent the risk of exceeding limits, under the condition of conventional new energy proportion. The preferred value is 0.7 to 0.8.

[0036] To visually demonstrate the ability of extreme value extrapolation to capture extreme error characteristics, this embodiment takes a high-proportion new energy scenario where wind and solar power account for more than 50% as an example. An empirical weighting parameter of 0.25 is set, and a comparative diagram of the probability density fitting of new energy prediction errors is drawn, as shown below. Figure 2 As shown. By Figure 2 It is known that, using the 95th percentile as the threshold for extreme values, the probability density of a traditional single KDE fitting curve rapidly decays and approaches zero after exceeding this threshold. This means that during scenario sampling, the system will consider the probability of extreme fluctuations to be zero, thus missing historical extreme error sample points that actually exist in the tail region. However, the proposed distribution constructed in this scheme obtains significant fat-tail probability compensation in the tail region. This effectively assigns historical extreme error sample points that were originally ignored by traditional methods a probability of occurrence that conforms to objective statistical laws. It not only retains the main statistical characteristics of the power grid's normal fluctuations, but also quantifies the real risk of low-probability extreme fluctuations, thus providing a reliable probabilistic basis for the subsequent generation of typical scenario sets.

[0037] 1.2.2 An initial scene set is generated by sampling based on the proposal distribution, and the importance weights of the initial scene set under the error distribution and the proposal distribution are calculated, as follows: Using the prediction error variable as a randomly generated variable, an inverse transformation sampling method is used to generate several random error time series as an initial scenario set, where each error time series corresponds to an initial scenario. The number of initial scenarios is derived based on the confidence level and error limit of the Monte Carlo simulation. For example, to ensure that the result has a 95% statistical confidence level and the lower limit of error is constrained to within 2%, the sampling number is 50,000. The specific steps for generating the error time series are as follows: the probability density function of the proposed distribution is discretely accumulated to obtain the corresponding cumulative distribution function; based on the randomly generated variable, random numbers uniformly distributed in the interval [0,1] are generated by the system random number generator; the uniformly distributed random numbers are substituted into the inverse function of the cumulative distribution function for inverse solution to obtain the corresponding error variable prediction value; finally, according to the preset system control time step (e.g., 15 minutes), the above steps of random number generation and inverse solution are repeated, and the error variable prediction values ​​corresponding to each time step are arranged in chronological order to obtain a discrete error time series.

[0038] To restore the objective probabilistic statistical characteristics of the real power grid and eliminate the dimensional misalignment between the discrete sampling space and the continuous fitting space, the system first uses kernel density estimation (KDE) to smooth the historical discrete real error data and fit a continuous baseline probability density function. Then, it calculates the ratio of the probability density of each initial scenario to the probability density of the proposed distribution under the baseline probability density function, defines it as an importance weight coefficient, and uses this coefficient to correct the probability of the initial scenario. The formula for calculating the importance weight is as follows: (6) in, For the first The importance weight coefficient of each initial scene; For the first The error time series vector of each initial scene; Error time series vector The probability density value is calculated by substituting it into the aforementioned baseline probability density function; Error time series vector Substitute into the proposed distribution formula The density function value obtained from the calculation.

[0039] 1.2.3 The probability of extreme scenarios in the initial scenario set is corrected using importance weights, and the corrected initial scenario set is clustered into a typical scenario set, as follows: The calculated importance weight coefficients are multiplied by the initial probability of occurrence of each corresponding initial scenario to obtain the corrected probability of occurrence. This is to offset the amplification effect of the proposed distribution on the probability of occurrence of extreme scenarios in the tail, so that it returns to the true low probability attribute, thereby correcting the probability of occurrence of extreme scenarios in the initial scenario set.

[0040] A scene reduction algorithm based on Wasserstein distance is adopted. Each initial scene in the initial scene set is regarded as a discrete point in the probability space, and its corrected occurrence probability is the probability mass of the point. The Euclidean distance between all pairs of scenes is calculated as the probability transport cost. Then, the scene pairs that reduce the Wasserstein distance the fastest are iteratively searched and merged. The regular scenes with very close distances are merged into one and the probability masses are added together until the number of retained scenes converges to 10 to 15, and finally the typical scene set is obtained.

[0041] 1.2.4 Extract the net load gap from the typical scenario set, analyze the peak and fluctuation characteristics of the net load gap, and map them to the adjustment demand vector, as follows: Extract the net load gap time series of each typical scenario in the typical scenario set, and use the local absolute extreme value of the time series as the "power demand" component in the preset regulation demand vector; perform point-by-point discrete comparison between the value of each time segment of the net load gap time series and the preset regulation rated baseline, count the number of time segments in which the net load gap value continuously exceeds the baseline, and sum the time steps of the consecutively exceeding segments as the "demand duration" component in the preset regulation demand vector. The regulation rated baseline can be set as the maximum allowable power deviation threshold specified in the power grid safe operation guidelines; use the physical time difference (in seconds) experienced by the power grid frequency dropping to the primary frequency regulation trigger dead zone caused by the sudden change in net load power as the "allowable delay" component of the underlying hardware action in the preset regulation demand vector. By vectorizing and combining the extracted values ​​of the three physical dimensions—power demand, demand duration, and allowable delay—a preset adjustment demand vector corresponding to each typical scenario can be obtained.

[0042] This step constructs a proposal distribution that includes tail extreme value extrapolation and uses importance weights for correction. This balances the effective coverage of low-probability extreme fluctuations in the power grid by scenario sampling with the distortion bias caused by extreme value sampling, objectively restoring the true uncertainty of the system. In addition, the net load gap is analyzed and mapped into multi-dimensional physical indicators of power, time and delay. The abstract system-level fluctuations are deconstructed into discrete operating parameters that can be directly identified by the underlying hardware. This realizes the quantitative transformation of abstract power grid fluctuations into executable constraints of the underlying hardware, laying a quantitative data foundation for the subsequent accurate comparison of multi-dimensional feasible domains.

[0043] 1.3 Extract the minimum regulation capacity constraints for each energy storage device, including: 1.3.1 Compare the preset adjustment demand vector with the aggregated feasible region, and extract the responsive adjustment demands that do not exceed the aggregated feasible region, as follows: First, the power demand in the preset adjustment demand vector is divided by the allowable delay (and time unit conversion is performed) to calculate the minimum equivalent ramp rate required to meet the response speed requirement. Then, the original preset adjustment demand vector is equivalently converted into an aligned demand vector containing "power demand, demand duration, and equivalent ramp rate". The formula for calculating the minimum equivalent ramp rate is as follows: (7) in, The minimum equivalent climbing rate (MW / min); The power demand (MW) in the preset demand vector is used for adjustment. Multiply the allowable delay (in seconds) in the preset adjustment demand vector by 60 to unify the time unit to minutes.

[0044] The power demand, demand duration, and equivalent ramp rate components in the aligned demand vector are compared numerically with the aggregated maximum charge / discharge power boundary, aggregated maximum continuous discharge time boundary, and aggregated maximum ramp rate boundary in the aggregated feasible region, respectively. If all components of the aligned demand vector are within the envelope of the boundaries in the aggregated feasible region, the entire aligned demand vector is directly extracted as a responsive and adjustable demand. If there are components in the aligned demand vector that exceed the boundaries of the aggregated feasible region, the excess portion is truncated to the corresponding maximum extreme value boundary, and the truncated safety values ​​are used to reconstruct the responsive and adjustable demand. For example, if the power demand in the aligned demand vector is 10MW, but the comparison reveals that the current maximum charge / discharge power boundary of the aggregated feasible region is only 8MW, then 8MW is directly truncated as the upper limit of the actual power allocation in the responsive and adjustable demand, thereby eliminating the false demand portion that the underlying hardware cannot execute.

[0045] 1.3.2 Based on the maximum charge / discharge power and depth of discharge of each energy storage device, the responsive adjustment demand is mapped to the corresponding target energy storage device, and this mapping serves as the minimum adjustment capability constraint for the target energy storage device, as detailed below: First, the specific component values ​​of the responsive and adjustable demand in each dimension (i.e., power, time, and ramp rate) of the aggregated feasible domain are extracted. These component values ​​are then compared with preset threshold values ​​for their respective dimensions to determine the dimensions that trigger the system demand exceeding limits. If two or three dimensions simultaneously exceed their corresponding preset threshold values, causing a conflict, the dimension with the highest priority is locked as the core exceeding limit dimension according to a preset priority order, such as "equivalent ramp rate is optimal, demand duration is secondary, and power demand is the least important." Based on the adjustment characteristics of the determined core exceeding limit dimension, reverse matching of physical characteristics is performed within the aggregated feasible domain to identify heterogeneous energy storage nodes with corresponding advantageous adjustment capabilities as target energy storage devices. For example, if the extracted equivalent ramp rate component is greater than the preset first extreme threshold (e.g., 500MW / min, indicating an overshoot of the extreme response limit), then power-type equipment with millisecond-level step response characteristics (e.g., high-speed flywheel energy storage) is prioritized as the target energy storage device; if the extracted equivalent ramp rate component is between the second extreme threshold and the first extreme threshold (e.g., 10MW / min to 500MW / min, indicating an overshoot of the normal ramp limit), then equipment with high-frequency power regulation capabilities (e.g., lithium-ion battery clusters) is prioritized as the target energy storage device; if the demand duration component is greater than the preset time threshold (e.g., 120min, indicating an overshoot of the energy limit), then energy-type equipment with long-term continuous discharge characteristics (e.g., vanadium redox flow batteries) is prioritized as the target energy storage device; if only the power demand component is greater than the preset power threshold (e.g., 20MW, indicating an overshoot of the pure active power amplitude), then energy-power balancing equipment with large-capacity basic power support capabilities (e.g., centralized lithium iron phosphate energy storage power stations) is prioritized as the target energy storage device.

[0046] Subsequently, the maximum charge / discharge power extreme value of each target energy storage device is multiplied by its remaining available discharge ratio (i.e., the factory-set maximum safe discharge depth minus the current measured discharge depth) to obtain the initial weight factor of each target energy storage device. The initial weight factors of all target energy storage devices are summed to obtain the total system weight, and the discrete proportional allocation coefficient of the initial weight factor of a single target energy storage device to the total system weight is calculated. Finally, the responsive adjustment demand is decomposed by dimension, and only the "power demand" component is multiplied by the discrete proportional allocation coefficient of each target energy storage device to obtain the active power reserve value mapped to the corresponding target energy storage device. This value serves as the minimum adjustment capability constraint for the target energy storage device, ensuring that the system has a definite physical response capability when dealing with future extreme operating conditions and avoiding the risk of exceeding limits when control commands are issued.

[0047] To verify the effectiveness of the anti-limit-overrun technology in step 1 of this invention, which generates the minimum regulation capability constraint, a comparative experiment was conducted based on a regional microgrid of a heterogeneous energy storage cluster. The experiment included a 5MW / 10MWh lithium iron phosphate energy storage unit A, a 2MW / 8MWh vanadium redox flow battery energy storage unit B, and a 1MW / 0.1MWh high-speed flywheel energy storage unit C. The preset regulation demand vector derived from the extreme scenario analysis was: power demand 4.2MW, demand duration 45min, and allowable delay 0.5s. Using a conventional proportional (5:2:1) allocation method, energy storage C was allocated a continuous output of 0.525MW. Although this value met the power dimension requirement, because the currently available energy of energy storage C was only 0.062MWh and the maximum discharge time was only 7.1min, due to the lack of multi-dimensional boundary verification, energy storage C ran out of energy and disconnected from the grid in the 8th minute, resulting in a secondary power drop of 0.525MW and regulation failure. In contrast, the present invention first converts the allowable delay of 0.5s into an equivalent ramp rate requirement to align the dimensions. Through comparison and aggregation of feasible domains, it identifies that while flywheel energy storage C meets the requirements for rapid ramping and power, it suffers from a "time dimension" mismatch. Subsequently, a cutoff mechanism is triggered, strictly limiting the minimum adjustment capability constraint of energy storage C to a safe physical boundary of 7.1 minutes. The remaining long-term power gap is dynamically mapped to energy storage devices A and B, which possess energy advantages. Simulation results show that under the control of the minimum adjustment capability constraint of this scheme, no equipment triggers over-limit protection throughout the entire process. The heterogeneous cluster safely fulfills the 4.2MW adjustment requirement, meeting the 0.5s response requirement and lasting for 45 minutes. This verifies that the present invention can effectively identify and intercept cross-dimensional physical shortcomings, fundamentally ensuring the physical executability of control commands.

[0048] This step will perform cross-dimensional comparison and discrete truncation between the extreme adjustment requirements based on extreme value extrapolation quantization and the three-dimensional aggregated feasible domain that integrates dynamic features, generating minimum adjustment capability constraints. By utilizing the physical boundary envelope of discrete polyhedra, cross-dimensional mismatch risks can be identified and eliminated in advance, overcoming the evaluation limitations caused by traditional reliance on one-dimensional static backup. Thus, at the scheduling source, instructions exceeding equipment limits are effectively avoided, reducing the risk of heterogeneous clusters going offline under complex and extreme conditions, and improving the security and executability of control instructions at the physical level.

[0049] Step 2: Construct a multi-timescale collaborative optimization model including a day-ahead scheduling layer, an intraday rolling layer, and a real-time control layer, and perform optimization solutions based on the minimum adjustment capability constraint.

[0050] 2.1 Construct a multi-timescale collaborative optimization model comprising a day-ahead scheduling layer, an intraday rolling layer, and a real-time control layer, including: It should be noted that the multi-timescale collaborative optimization model is specifically a pre-deduction architecture based on multi-stage stochastic programming, executed in the day-ahead phase. In this architecture, the day-ahead scheduling layer is used to determine the baseline operating plan issued by the current system; while the intraday rolling layer and real-time control layer are based on the typical scenario set generated in step 1 to simulate and predict possible future operating states. Its core is to simulate the physical response cost of the system in mitigating power disturbances under extreme fluctuation conditions in the future by pushing forward the discrete time step, and quantify the cost as the future deviation penalty cost, which is used as a feedback dual signal to reverse correct the day-ahead baseline plan, thereby generating feedforward control commands with strong disturbance resistance robustness in the day-ahead phase.

[0051] 2.1.1 Configure the optimization model for the day-ahead scheduling layer, as follows: Decision variables: Start-up and shutdown status of slow thermal power or large hydropower units and basic planned active power output curves at discrete time steps; The formula for the objective function is as follows: (8) Unit fuel consumption cost Costs of unit start-up and shutdown operations The formula is as follows: (9) (10) in, The objective function value of the day-ahead scheduling layer; This represents the total number of discrete time steps within the current day's scheduling cycle. This is the index of the discrete day-ahead time step; For slow-speed units, the day-ahead time step The basic plan has active power output variables (megawatts, MW); To characterize the day-ahead time step of the slow-speed unit A 0-1 integer variable representing the start / stop status; This is the inherent fuel consumption coefficient of the generator unit. The day-ahead time step (e.g., 1 hour); , These are the cost constants for a single start-up and shutdown, respectively; That is, the aforementioned day-ahead comprehensive operating cost; This represents the total number of typical scenarios; For the first The probability density of occurrence of a typical scenario; In the first The deviation penalty cost of intraday rolling layer feedback in a typical scenario. This cost term physically represents the additional economic cost that the underlying system is expected to bear in order to cope with the fluctuations in this scenario. As a dynamic transfer variable connecting the day-ahead and intraday layers, it is assigned a value of zero in the initial iteration and its value increases dynamically in subsequent solutions as the degree of the underlying layer exceeds the limit.

[0052] 2.1.2 Configure the optimized model for the intraday rolling layer, as follows: Decision variables: the active power adjustment of continuous charging and discharging of each heterogeneous energy storage device, and the active power output adjustment of fast-response units (such as gas turbines); The formula for the objective function is as follows: (11) Intraday dispatching and operating costs (i.e., fuel costs for rapid unit response) The formula is as follows: (12) in, Typical scenario The objective function value of the intraday rolling layer below; This represents the number of intraday discrete time steps contained within a single daytime step, where the intraday discrete time step can be set to 15 minutes. In the scene Next day's time step Decision variables (MW) for active power adjustment of internal high-speed generating units and energy storage output. , For fast-acting units, The time step within the day (unit converted to hours); The recommended value for the day-ahead tracking deviation penalty coefficient is 500 to 1000 (RMB / MW). The cost of tracking deviations in the aforementioned day-ahead plan; In the scene The deviation penalty cost fed back by the real-time control layer represents the extreme balancing cost that the underlying hardware is expected to bear in order to cope with power imbalance jumps in a short timescale under extreme operating conditions. It also serves as a dual variable for cross-layer dynamic transmission, feeding back the degree of exceeding the limit at the real-time level to the intraday and day-ahead layers.

[0053] 2.1.3 Configure the optimization model for the real-time control layer, as follows: Decision variables: Active power following allocation instructions from Automatic Generation Control (AGC), and the amount of wind and solar curtailment and load shedding that are forced to be triggered within the underlying hardware response dead zone; The formula for the objective function is as follows: (13) Energy storage regulation costs and real-time power tracking error cost The formula is as follows: (14) (15) in, In typical scenarios The objective function value of the lower real-time control layer; This represents the number of real-time discrete time steps contained within a single intraday time step, where the real-time discrete time step size can be set to 1 minute. In the scene Next real-time step The second-level active power tracking adjustment (MW) allocated to hardware devices. The depreciation constant per megawatt of energy storage equipment lifespan (RMB / megawatt); For real-time time steps The amount of load shedding or abandonment of active power from new energy sources (MW) is forcibly triggered by insufficient internal physical response. A high penalty constant for default (e.g., 10,000 yuan / megawatt-hour). This is the real-time time step (unit converted to hours).

[0054] It should be noted that the above The value corresponds to the power component difference truncated in step 1.3.1; specifically, when the component values ​​in the alignment demand vector generated in extreme scenarios cannot be completely within the envelope of the boundary of the aggregate feasible region, the system will truncate the power values ​​in the alignment demand vector that exceed the boundary of the aggregate feasible region; in the discrete optimization of the real-time control layer, the system directly assigns the truncated and discarded power difference value as the real-time time step. within This difference, exceeding hardware limits, cannot be realized by the underlying physical capabilities. In the model's decision-making process, it is forcibly and equivalently replaced by the corresponding load shedding or the abandonment of active power from renewable energy sources, and thus... Introducing tracking deviation penalty costs serves as a reverse constraint to ensure sufficient physical margin for upper-level scheduling.

[0055] 2.2 Optimization solution based on the minimum adjustment capability constraint, including: This embodiment employs the Alternating Directional Multiplier Method (ADMM) as the solution framework, establishing a cross-layer iterative mechanism within a day-ahead decision cycle that "transmits boundary constraints from top to bottom and provides feedback on penalty costs from bottom to top." Specifically, the system takes minimizing the objective function of the day-ahead scheduling layer as the global optimization objective, and uses the minimum regulation capacity constraint of the target energy storage device generated in step 1 as the global hard physical boundary. In each iteration of optimization, if the generated day-ahead basic planned output curve violates this boundary, its corresponding optimization branch is directly pruned and removed.

[0056] The specific solution steps are as follows: 2.2.1 The day-ahead reserve margin generated by the day-ahead scheduling layer is used as the boundary constraint of the intraday rolling layer, and the intraday power feasible region generated by the intraday rolling layer is used as the boundary constraint of the real-time control layer, as detailed below: In a single iteration of top-down forward extrapolation, the system first solves the day-ahead scheduling layer model to obtain the optimal solution for the basic planned active power output curve of the slow-speed units in the current iteration. It then calculates the difference between this optimal solution and the physical limits of the unit's maximum / minimum output on its nameplate, defining the extreme values ​​of the uplink and downlink reserve capacity formed by this difference as the day-ahead reserve margin, which is then sent to the intraday rolling layer as a rigid boundary constraint. Subsequently, based on the received day-ahead reserve margin, the intraday rolling layer model is optimized to obtain the optimal solution for the intraday adjustment plan, including the adjustment amounts for heterogeneous energy storage charging and discharging and fast-speed units. Combining this with the real-time state of charge of the energy storage devices, the upper and lower limits of the maximum instantaneous power adjustment that the underlying system can support at each time step are calculated, collectively referred to as the intraday power feasible region, and sent to the real-time control layer as the rigid feasible region boundary.

[0057] This step addresses the problem in existing multi-timescale scheduling models where unidirectional open-loop dispatching and static fixed backup configurations cause upper-level scheduling instructions to deviate from the actual response capabilities of the lower-level system, easily leading to unit over-limit disconnection during the real-time phase. It ensures the strict physical nesting of the multi-timescale decision space, thereby effectively preventing the risk of cascading disconnection triggered by unit over-limit during the real-time phase.

[0058] 2.2.2 Calculate the deviation penalty cost between the intraday rolling layer and the real-time control layer under the corresponding boundary constraints, as follows: 2.2.2.1 Obtain the deviation penalty cost of the real-time control layer Extract the continuous charging and discharging active power adjustment of heterogeneous energy storage devices and the active power output adjustment of fast response units from the daily rolling layer optimization output of the current cycle. The total power of the two is superimposed as the reference power trajectory sent to the real-time layer. Subsequently, calculate the power deviation between the real-time control layer's decision output and the reference power trajectory. The decision output is specifically the active power following command value actually called by the real-time control layer to smooth out high-frequency fluctuations.

[0059] After solving the real-time control layer model for the current round, since the intraday feasible region issued by the intraday layer is represented as a set of linear inequality constraints in the mathematical model, the dual variables can be obtained simultaneously using the dual solution function of a conventional optimization solver (such as a linear solution algorithm based on the interior point method or simplex method). Specifically, while obtaining the optimal solution of the real-time control layer objective function, based on the complementary relaxation theorem in the KKT conditions, the Lagrange multipliers (i.e., dual variables or shadow prices) bound to the upper or lower bound inequality constraints of the intraday power feasible region are directly read through the solver's bottom-level interface and used as the sensitivity coefficient. The Lagrange multipliers quantify the marginal value of the physical boundary, i.e., the optimal marginal rate of change of the real-time control layer objective function when the intraday feasible region boundary relaxes outward or tightens inward by one physical unit. The product of the power deviation and the sensitivity coefficient is used as the deviation penalty cost of the real-time control layer. The calculation formulas for the sensitivity coefficient and the deviation penalty cost of the real-time control layer are as follows: (16) (17) in, The extracted Lagrange multipliers, i.e., the sensitivity coefficients (yuan / megawatt); This refers to the Lagrange relaxation function corresponding to the real-time control layer. The upper or lower limit of the feasible power domain during the day is the value encountered when the real-time control layer is optimizing. When the upper or lower limit is not touched, the Lagrange multiplier is zero. The cost of penalty for deviations in the updated real-time control layer; The power deviation (MW) between the power output of the real-time control layer and the reference power trajectory is determined.

[0060] 2.2.2.2 Obtain the deviation penalty cost of the intraday rolling layer The net load deficit time series of a typical scenario is algebraically subtracted from the current period's day-ahead baseline planned output curve at the corresponding time cross-section. The resulting power residual time series represents the intraday source load forecast power demand that needs to be smoothed out by intraday and real-time layers. Next, the intraday source load forecast power demand is compared with the day-ahead reserve margin. If the forecast power demand is greater than the day-ahead reserve margin, it indicates that the day-ahead plan is too aggressive, and the absolute value of the difference is taken as the reserve capacity deficit; conversely, if the margin is sufficient, the deficit is zero.

[0061] The proportion of the reserve capacity deficit to the initial range of the intraday power feasible domain is calculated, and the deviation penalty cost of the real-time control layer is nonlinearly weighted to obtain the deviation penalty cost of the intraday rolling layer, as shown in the following formula: (18) in, This is the updated intraday rolling layer deviation penalty cost; For the natural constant An exponential function with base 0; The preset penalty index scaling factor can be set to 3.0 to adjust the steepness of the risk outbreak. Reserve capacity deficit (MW); and These are the upper and lower limits (MW) of the initial range of the feasible power domain during the day.

[0062] 2.2.3 The deviation penalty costs between the intraday rolling layer and the real-time control layer are respectively inversely added to the objective functions of the day-ahead scheduling layer and the intraday rolling layer for iterative solution, as follows: The deviation penalty cost calculated for each layer in the current round (i.e. and This feedback term is added to the day-ahead and day-intraday objective functions for the next round of iterative calculation. The system iteratively executes top-down boundary propagation and bottom-up penalty feedback until convergence and a cooperative control command is issued. The convergence determination requires simultaneous satisfaction of the objective function convergence condition and the residual tolerance condition. The specific convergence determination conditions are as follows: (19) (20) in, This represents the current iteration round of the ADMM algorithm; and These are the objective function values ​​of the day-ahead scheduling layer for the two iterations. The preset convergence accuracy limit is recommended to be 10. -2 ; For the first The original residual L2 norm at the next iteration specifically represents the L2 norm of the algebraic difference between the cross-layer transfer parameters (including day-ahead reserve margin, baseline power trajectory and intraday power feasible domain boundary extreme value) output by the upper layer optimization in the current round and the corresponding parameter values ​​actually received by the lower layer during local solution. For the first The dual residual L2 norm at the next iteration specifically represents the L2 norm of the algebraic difference between the calculation result of the cross-layer transfer parameter in the current iteration of the lower-layer model and the calculation result of the previous iteration. and These are the preset original residual tolerance and dual residual tolerance, respectively; a value of 10 is recommended for both. -4 .

[0063] To intuitively verify the convergence stability of the multi-timescale collaborative optimization model when dealing with the extreme constraints of heterogeneous hardware, this embodiment simulates the iterative optimization process of the ADMM algorithm; such as Figure 3 As shown in the figure, this is a schematic diagram of the convergence trajectory of the ADMM algorithm for collaborative optimization. Figure 3 As can be seen, in the initial optimization stage, due to the constraint of the hard physical boundary across layers, there is a significant tug-of-war and oscillation between the original residual and the dual residual. Around the 16th iteration, the algorithm triggers an adaptive update jump of the penalty parameter, which effectively suppresses the divergence of the original residual while causing a brief surge in the dual residual. As the cross-layer penalty transmission mechanism continues to take effect, the two types of residuals show a rapid logarithmic decrease after experiencing local conflicts, and finally fall into the convergence tolerance threshold line simultaneously at the 32nd iteration. This simulation trajectory fully demonstrates that the cross-layer closed-loop iterative mechanism constructed in this embodiment has excellent numerical convergence and engineering implementation capabilities.

[0064] 2.3 After the model iterative convergence, cooperative control commands are generated and issued, including: When the convergence criterion is met, the multi-timescale collaborative optimization model is determined to have ended its optimization process; the optimal decision variables of each level in the current convergence state are extracted and reorganized according to the execution sequence to generate the collaborative control command for issuance.

[0065] The coordinated control instructions specifically include: A rigid execution sequence: Extracting the optimal decision solution from the day-ahead scheduling layer, including the discrete start-stop states of slow-speed thermal power or large hydropower units and the basic planned active power output curves, and using this as a benchmark control instruction adapted to the current actual physical operation stage, directly issuing it to the underlying physical equipment for execution. A dynamic contingency plan sequence: Addressing the uncertainty and randomness of future grid operation, extracting the simulated optimal solutions generated by the intraday rolling layer and real-time control layer under various typical scenarios, such as continuous adjustment amounts of various heterogeneous energy storage devices, output adjustment amounts of fast-response units, second-level active power following allocation instructions, and wind and solar curtailment and load shedding amounts under extreme over-limit conditions; the above simulated decision variables are mapped and assembled into an emergency contingency plan library to cope with different extreme fluctuation conditions, and issued together with the rigid execution sequence to the underlying dispatch automation system (such as the EMS system) for online standby.

[0066] To verify the performance advantages of the multi-timescale collaborative optimization model proposed in this invention in dealing with continuous extreme operating conditions, a simulation environment was built based on the actual operating data of a typical regional power grid (including 50MW wind power, 30MW photovoltaic and 15MW / 30MWh energy storage cluster). Real source-load data from three consecutive days were selected as the test set, including a continuous extreme test scenario where photovoltaic output plummeted by 40% on the second day and wind power output exhibited severe heavy-tailed fluctuations on the third day. Existing technologies employ conventional multi-scale rolling optimization scheduling strategies, performing unidirectional open-loop feedback based on physical power deviations between time scales, and using a fixed proportion (e.g., 10%) of spinning reserve. Because they cannot detect and verify the physical constraints of the underlying energy storage cluster in advance, the control command power repeatedly exceeds the equipment limits and triggers equipment disconnection from the grid, ultimately resulting in an 8.5MWh active power deficit and high penalty costs. This invention introduces a minimum adjustment capability constraint to prune unexecutable demand components in advance. Furthermore, in cross-layer iterations, a deviation penalty transmission mechanism is used to quantify short-term physical exceedance risks as economic costs embedded in the upper-layer objective function. After sensing the extreme risks on the second and third days, the upper-layer scheduling model adaptively expands the day-ahead safety margin to 18.5%. Experimental data shows that this invention avoids equipment exceeding limits and disconnecting from the network, reduces the total active power deficit by 85.9%, and lowers the total operating cost by 13.7%. These results demonstrate that this invention overcomes the shortcomings of traditional unidirectional open-loop scheduling, which is prone to detachment from the actual execution capabilities of the underlying layers, and improves the system's robustness against disturbances and overall operational economy.

[0067] This step constructs a multi-timescale collaborative optimization model covering day-ahead, intraday, and real-time scenarios. It uses the minimum adjustment capability constraint as the physical boundary to prune the optimization branches and finally extracts the optimal decision variables to generate collaborative control commands after iterative convergence. By using the extreme values ​​of the actual response of the underlying devices to constrain the upper-level economic scheduling, it overcomes the technical defect of traditional unidirectional scheduling commands being prone to deviating from the actual execution capabilities of the underlying devices. This achieves cross-timescale collaborative regulation, ensures the physical executability of the issued collaborative control commands, and improves the robustness of the system under complex operating conditions.

[0068] Example 2, Figure 4 A source-load-storage coordinated optimization control system for new energy entities is presented, comprising the following modules: The constraint generation module is used to construct an aggregated feasible region based on the operating status and multi-dimensional adjustment capabilities of energy storage devices, and extract the minimum adjustment capability constraints of each energy storage device in combination with the preset adjustment demand vector. The collaborative optimization module is used to construct a multi-timescale collaborative optimization model including a day-ahead scheduling layer, an intraday rolling layer, and a real-time control layer, and to perform optimization solutions based on the minimum adjustment capability constraint. During the solution process, the daytime reserve margin generated by the daytime scheduling layer is used as the boundary constraint of the intraday rolling layer, and the intraday power feasible region generated by the intraday rolling layer is used as the boundary constraint of the real-time control layer. Calculate the deviation penalty cost between the intraday rolling layer and the real-time control layer under the corresponding boundary constraints, and then add it back to the objective function of the day-ahead scheduling layer and the intraday rolling layer respectively. Iterate until convergence, and then issue a coordinated control command.

[0069] The above formulas are all dimensionless calculations. The formulas are derived from software simulations using a large amount of collected data, and are the closest to the real situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0070] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0071] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0072] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0073] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0074] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A source-load-storage coordinated optimization control method for new energy entities, characterized in that, Includes the following steps: An aggregated feasible region is constructed based on the operating status and multi-dimensional adjustment capabilities of energy storage devices, and the minimum adjustment capability constraints of each energy storage device are extracted by combining the preset adjustment demand vector. A multi-timescale collaborative optimization model is constructed, comprising a day-ahead scheduling layer, an intraday rolling layer, and a real-time control layer, and optimization is performed based on the minimum adjustment capability constraint. During the solution process, the daytime reserve margin generated by the daytime scheduling layer is used as the boundary constraint of the intraday rolling layer, and the intraday power feasible region generated by the intraday rolling layer is used as the boundary constraint of the real-time control layer. Calculate the deviation penalty cost between the intraday rolling layer and the real-time control layer under the corresponding boundary constraints, and then add them in reverse to the objective functions of the day-ahead scheduling layer and the intraday rolling layer, respectively. Iterate until convergence, and then issue a cooperative control command. The calculation of the deviation penalty cost between the intraday rolling layer and the real-time control layer under corresponding boundary constraints includes: extracting the power deviation between the baseline power trajectory issued by the intraday rolling layer to the real-time control layer and the decision output of the real-time control layer; using the dual variable corresponding to the boundary of the intraday power feasible region as a sensitivity coefficient and multiplying it with the power deviation to obtain the deviation penalty cost of the real-time control layer; calculating the reserve capacity deficit based on the intraday source load predicted power demand and the day-ahead reserve margin; and performing a nonlinear weighting of the deviation penalty cost of the real-time control layer based on an exponential function according to the proportion of the reserve capacity deficit to the initial range of the intraday power feasible region to obtain the deviation penalty cost of the intraday rolling layer.

2. The method according to claim 1, characterized in that, The components of the preset adjustment demand vector include power demand, demand duration, and allowable delay; the steps for obtaining the adjustment demand vector include: Extreme value extrapolation fitting and probability density estimation are performed on the tail and main body of the error distribution in the historical prediction error data of new energy, respectively, and combined to obtain a proposed distribution for scenario sampling. An initial scene set is generated by sampling based on the proposed distribution, and a typical scene set is obtained by clustering. Extract the net load gap of the typical scenario set, analyze the peak and fluctuation characteristics of the net load gap, and map them into the adjustment demand vector.

3. The method according to claim 2, characterized in that, The clustering yields a set of typical scenarios, including: Calculate the importance weights of the initial scene set under the error distribution and the proposal distribution; Using the importance weights, the probability of extreme scenarios occurring in the initial scenario set is corrected, and the corrected initial scenario set is clustered into the typical scenario set.

4. The method according to claim 1, characterized in that, The construction of the aggregated feasible domain based on the operating status and multi-dimensional adjustment capabilities of energy storage devices includes: The initial state of charge of each energy storage device in the current cycle is obtained, and the charging and discharging power boundary, continuous discharge time and dynamic response characteristics of each energy storage device are combined to construct the device feasible domain including the dimensions of power, time and ramp rate. The feasible regions of each device are aggregated to obtain the aggregated feasible region.

5. The method according to claim 1, characterized in that, The extraction of the minimum regulation capability constraints for each energy storage device includes: Compare the preset adjustment demand vector with the aggregated feasible region, and extract the responsive adjustment demand that does not exceed the aggregated feasible region; Based on the maximum charge / discharge power and depth of discharge of each energy storage device, the responsive adjustment requirements are mapped to the corresponding target energy storage device, and used as the minimum adjustment capability constraint of the target energy storage device.

6. The method according to claim 5, characterized in that, The selection steps for the target energy storage device include: Extract the component values ​​of the responsive and adjustable demand in each dimension of the aggregated feasible domain, and compare them with the corresponding preset judgment thresholds to obtain the dimensions that exceed the limit; When multiple limits exist, the core limit-breaking dimension is determined according to a preset priority order, and the corresponding target energy storage device is matched based on the adjustment characteristics of the core limit-breaking dimension.

7. The method according to claim 1, characterized in that, The objective function of the day-ahead scheduling layer includes the day-ahead comprehensive operating cost and the deviation penalty cost of the intraday rolling layer; the objective function of the intraday rolling layer includes the intraday scheduling operating cost, the day-ahead plan tracking deviation cost, and the deviation penalty cost of the real-time control layer; the objective function of the real-time control layer includes the real-time power tracking deviation cost and the energy storage regulation cost.

8. A source-load-storage coordinated optimization control system for new energy entities, characterized in that, Includes the following modules: The constraint generation module is used to construct an aggregated feasible region based on the operating status and multi-dimensional adjustment capabilities of energy storage devices, and extract the minimum adjustment capability constraints of each energy storage device in combination with the preset adjustment demand vector. The collaborative optimization module is used to construct a multi-timescale collaborative optimization model including a day-ahead scheduling layer, an intraday rolling layer, and a real-time control layer, and to perform optimization solutions based on the minimum adjustment capability constraint. During the solution process, the daytime reserve margin generated by the daytime scheduling layer is used as the boundary constraint of the intraday rolling layer, and the intraday power feasible region generated by the intraday rolling layer is used as the boundary constraint of the real-time control layer. The deviation penalty cost of the intraday rolling layer and the real-time control layer under the corresponding boundary constraints is calculated and then inversely added to the objective functions of the daytime scheduling layer and the intraday rolling layer, respectively. The solution is iterated until convergence, and then the cooperative control command is issued. The calculation of the deviation penalty cost between the intraday rolling layer and the real-time control layer under corresponding boundary constraints includes: extracting the power deviation between the baseline power trajectory issued by the intraday rolling layer to the real-time control layer and the decision output of the real-time control layer; using the dual variable corresponding to the boundary of the intraday power feasible region as a sensitivity coefficient and multiplying it with the power deviation to obtain the deviation penalty cost of the real-time control layer; calculating the reserve capacity deficit based on the intraday source load predicted power demand and the day-ahead reserve margin; and performing a nonlinear weighting of the deviation penalty cost of the real-time control layer based on an exponential function according to the proportion of the reserve capacity deficit to the initial range of the intraday power feasible region to obtain the deviation penalty cost of the intraday rolling layer.

Images