A virtual power plant self-adapting scheduling method based on health degree credible derating

Abstract

This application relates to the technical field of virtual power plant scheduling optimization, and in particular to an adaptive scheduling method for virtual power plants based on health-based reliability derating. The method includes: acquiring resource information and external scheduling instructions corresponding to multiple adjustable resources in the virtual power plant; modeling resource capabilities and constraints based on the resource information to obtain a basic feasible region; calculating the corresponding health score based on the communication status and historical performance deviation of each adjustable resource; performing reliability derating on the basic feasible region based on several health scores to obtain a committable feasible region; calculating an adaptive margin based on the health scores and prediction error levels of all adjustable resources; reducing the committable feasible region based on the adaptive margin to obtain an executable feasible region; and constructing a rolling scheduling optimization based on external scheduling instructions and the executable feasible region to obtain and execute a scheduling scheme for each adjustable resource. This application has the effect of optimizing the accuracy of reliable commitment capabilities to improve the performance rate while also considering benefits.

Description

Technical Field

[0001] This application relates to the technical field of virtual power plant scheduling optimization, and in particular to an adaptive scheduling method for virtual power plants based on health-based reliable derating. Background Technology

[0002] Virtual power plants need to "commit" adjustable capacity / power (committing to increasing or decreasing dispatch capacity), but resource-side issues often arise such as equipment status fluctuations, communication packet loss and latency, temporary user disconnection, and inconsistent control responses. Current methods often make commitments based on equipment rated capacity or static boundaries, leading to the following problems:

[0003] Over-commitment: Actual switchover or response deviations and penalties;

[0004] Overly conservative: limiting profit reduction in order to avoid default;

[0005] Lack of dynamic credibility: Unable to adjust commitments and scheduling margins as implementation quality changes. Summary of the Invention

[0006] To optimize the accuracy of credible commitment capabilities to improve fulfillment rates while also considering benefits, this application provides an adaptive scheduling method for virtual power plants based on health-based credible derating.

[0007] This application provides an adaptive scheduling method for virtual power plants based on health-based reliable derating, employing the following technical solution:

[0008] An adaptive scheduling method for virtual power plants based on health-based reliable derating includes the following steps:

[0009] Obtain resource information and external dispatch instructions corresponding to multiple adjustable resources in the virtual power plant;

[0010] Based on the resource information, resource capacity and constraint modeling is performed to obtain a basic feasible region including the first up and down power boundary.

[0011] The corresponding health score is calculated based on the communication status and historical performance deviation of each adjustable resource.

[0012] Based on several of the aforementioned health scores, the basic feasible region is reliably reduced to obtain a committed feasible region that includes the second up and down power boundaries.

[0013] An adaptive margin is calculated based on the health score and prediction error level corresponding to all the adjustable resources. The feasible region is then reduced based on the adaptive margin to obtain an executable feasible region that includes the third up and down power boundary.

[0014] Based on the external scheduling instructions and the feasible execution domain, a rolling scheduling optimization is constructed to obtain and execute a scheduling scheme for each adjustable resource.

[0015] In some embodiments, the resource information includes the online status, maximum up-adjustment capacity, maximum down-adjustment capacity, ramp rate, energy information, and user service information of each adjustable resource. Based on this resource information, resource capacity and constraint modeling is performed to obtain a basic feasible region including a first up- and down-adjustment power boundary, comprising the following steps:

[0016] The first up-down power boundary is generated based on the maximum up-adjustment capability and the maximum down-adjustment capability combined with the online status.

[0017] Based on the ramp rate, a ramp rate constraint is generated. When the ramp rate is greater than 0, the absolute difference of the resource adjustment amount between adjacent time points is less than the ramp rate.

[0018] Based on the energy information, an energy constraint is generated such that when the resource adjustment amount is increased, the corresponding real-time energy decreases and the real-time energy constraint is within a preset energy capacity boundary.

[0019] Business constraints are generated based on the user business information to ensure that the function value can be calculated and is not greater than 0 after substituting the given resource adjustment amount, the real-time energy and the external business volume into the explicit constraint expression.

[0020] The resource adjustment amount that satisfies the first power adjustment boundary and all constraints is integrated into the basic feasible domain.

[0021] In some embodiments, a health score is calculated based on the communication status and historical performance deviations of each of the adjustable resources, including the following steps:

[0022] The system acquires online rate, command receipt success rate, packet loss rate, and round-trip latency, performs scaling and normalization, and calculates the communication score at each time point by combining the communication weight set.

[0023] The deviation amplitude of the target adjustment amount and the actual adjustment amount of each adjustable resource at the same time is obtained and divided by the reference power base to calculate the relative error. The response time of each adjustable resource to the command is obtained. The relative error and the response time are normalized by the exponential decay normalization function to obtain the error sub-score and the time sub-score. The alarm index of the adjustable resource at each time is obtained and normalized to generate an alarm score.

[0024] The performance score is calculated based on the error sub-score, the time sub-score, and the alarm score, combined with the corresponding performance weight set.

[0025] The health score is obtained by weighting and fusing the communication score and the performance score, and then using an exponentially weighted moving average for smooth updates.

[0026] In some embodiments, the underlying feasible region is reliably derated based on several of the said health scores to obtain a committable feasible region including a second up-and-down power boundary, including the following steps:

[0027] The calculated health score is used as a confidence coefficient, and the first up and down power boundary is multiplied by the confidence coefficient to calculate the second up and down power boundary.

[0028] In some embodiments, an adaptive margin is calculated based on the health score and prediction error level corresponding to all the adjustable resources, including the following steps:

[0029] A weighted weight is generated based on the maximum adjustment capability of each adjustable resource, and the health score is combined with the weighted weight to calculate the aggregate health score.

[0030] The aggregate tracking error is calculated by combining the measured power response data from the previous moment with the external scheduling command from the previous moment and the preset aggregate reference power benchmark.

[0031] Obtain the smoothing parameters and combine them with the aggregated tracking error. Calculate the prediction error variance at the current time using the exponentially weighted moving average method, and take the square root of the prediction error variance to calculate the prediction error level.

[0032] The adaptive margin is obtained by multiplying the aggregated health level and the prediction error level by the equilibrium coefficient and summing them.

[0033] In some embodiments, the following steps are also included:

[0034] The current operating scenario is obtained. If the current operating scenario is a short-term fluctuation, the equilibrium coefficient is increased; if the current operating scenario is a long-term reliable one, the equilibrium coefficient is decreased.

[0035] The larger the equilibrium coefficient, the more the change in the adaptive margin depends on the change in the prediction error level; the smaller the equilibrium coefficient, the more the change in the adaptive margin depends on the change in the aggregate health.

[0036] In some embodiments, reducing the committable feasible region based on the adaptive margin to obtain an executable feasible region including a third up-and-down power boundary includes the following steps:

[0037] The safety margin coefficient is calculated using the adaptive margin, and the second up and down power adjustment boundary is multiplied by the safety margin coefficient to calculate the third up and down power adjustment boundary.

[0038] In some embodiments, a rolling scheduling optimization is constructed based on the external scheduling instructions and the feasible execution domain to obtain and execute a scheduling scheme for each adjustable resource, including the following steps:

[0039] Configure the window length and the time period that needs to be optimized for each scheduling moment;

[0040] The resource adjustment amount of each adjustable resource in each time period within the rolling window is used as a decision variable;

[0041] The objective function of the quadratic programming is constructed with the goal of minimizing the overall cost. The objective function includes a quadratic term for the aggregate tracking error and a linear term for the resource usage cost. The quadratic term also includes a tracking priority weight, the larger the value of which indicates that the system has a higher requirement for tracking accuracy. The linear term also includes a usage coefficient, the larger the value of which indicates that the system is more inclined to avoid calling the adjustable resource.

[0042] Within the feasible region, the optimal decision solution that minimizes the objective function is obtained to obtain the optimal adjustment sequence for each time period within the rolling window;

[0043] Only the optimal adjustment value corresponding to the first time period in the optimal adjustment value sequence is executed and distributed to the corresponding adjustable resource.

[0044] In some of these embodiments, it also includes:

[0045] When any of the adjustable resources is detected to meet the preset abnormal triggering conditions, the health score is updated using the fast decay parameter, and the calculation of the second power adjustment boundary, the calculation of the adaptive margin, and the solution of the rolling scheduling optimization are re-executed.

[0046] In some of these embodiments, it also includes:

[0047] If the health score is lower than the removal threshold, then the online status of the adjustable resource is set to 0 and it is removed.

[0048] After removing the virtual power plants, the maximum upward adjustment capacity and the maximum downward adjustment capacity corresponding to them are recalculated.

[0049] If the scheduling scheme cannot satisfy the external scheduling instruction, then the external commitment is downgraded to generate a new external scheduling instruction to limit it to the range of aggregated available capabilities.

[0050] The technical solutions provided by the embodiments of this application have the following technical effects:

[0051] It balances operational reliability and economy from the root, significantly reducing default risk while maximizing dispatch benefits; it achieves standardized aggregation and refined management of distributed resources, and can automatically reconstruct dispatch schemes in abnormal scenarios, significantly improving operational robustness; the entire process is computable and easy to implement in engineering, adapting to the real-time dispatch requirements of the power market, effectively enhancing the grid friendliness and market competitiveness of virtual power plants. Attached Figure Description

[0052] Figure 1 This is a schematic diagram illustrating the steps of a virtual power plant adaptive scheduling method based on health-based reliable derating provided in this embodiment. Detailed Implementation

[0053] To better understand the purpose, technical solutions, and advantages of this application, it has been described and illustrated below with reference to the accompanying drawings and embodiments. However, those skilled in the art should understand that this application can be implemented without these details. In some cases, to avoid obscuring various aspects of this application due to unnecessary description, well-known methods, processes, systems, components, and / or circuits already described at a higher level will not be elaborated upon. It will be apparent to those skilled in the art that various modifications can be made to the embodiments disclosed in this application, and the general principles defined in this application can be applied to other embodiments and application scenarios without departing from the principles and scope of this application. Therefore, this application is not limited to the illustrated embodiments, but conforms to the broadest scope consistent with the scope of protection claimed in this application.

[0054] It should be noted that the descriptions of these embodiments are for the purpose of aiding understanding the present invention, but do not constitute a limitation thereof. Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0055] In the description of this application, "several" means one or more, "more than" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0056] In the description of this application, the terms "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples.

[0057] like Figure 1 As shown in the figure, this application discloses an adaptive scheduling method for virtual power plants based on health-reliable derating, which specifically includes the following steps:

[0058] S100 obtains resource information and external dispatch instructions corresponding to multiple adjustable resources in the virtual power plant.

[0059] Before proceeding with the steps outlined in this application, it is necessary to standardize discrete time and notation, which mainly includes:

[0060] Time dimension standardization. First, determine the discrete time step of the entire scheduling system. (In seconds or minutes), the continuous running time is divided into discrete scheduling time points k with a fixed step size, where k = 0, 1, 2... The status acquisition, command issuance, and scheduling solution of all adjustable resources are strictly aligned with this discrete time scale to ensure system-wide time synchronization.

[0061] Standardize resource and instruction dimensions. Assign a unique index number to all distributed resources within the virtual power plant. Clearly define the aggregated external dispatch instructions as (Unit: kW, positive indicates increase / increase in power generation, negative indicates decrease / reduction in power generation), unifying the direction and definition of power regulation across the entire system.

[0062] Standardize the decomposition of adjustment quantities. Determine the resource adjustment quantity for each adjustable resource. Split into upward components and reduce the amount The constraint requires that both components be non-negative, where This eliminates the constraint bias caused by the mixed calculation of positive and negative values, and provides a standardized calculation basis for subsequent power boundary constraints.

[0063] Resource online status standardization: Real-time collection of online identifiers for adjustable resources through device heartbeat and communication connection status. The online status is assigned a value of 1, and the offline status is assigned a value of 0. This flag will directly affect subsequent power boundary constraints to achieve automatic removal of offline resources.

[0064] After the above unification, the system can directly obtain resource information and external dispatch instructions corresponding to each adjustable resource in the virtual power plant. The resource information mainly includes the online status, maximum up-adjustment capacity, maximum down-adjustment capacity, ramp rate, energy information, and user service information of each adjustable resource.

[0065] S200, based on resource information, performs resource capability and constraint modeling to obtain the basic feasible region including the first up and down power boundary.

[0066] For each adjustable resource i, establish the corresponding set of adjustable capabilities at time k and define it as the basic feasible region. The basic feasible region includes the basic upper / lower power adjustment boundaries, corresponding to the upper and lower limits of an adjustment amount. Only when it falls within the range of the adjustment amount can the adjustment be considered feasible. Within range Only those resource-compliant and effective adjustment instructions can be executed; anything beyond that... Scope Such instructions would exceed the physical limits of the equipment and the rules of business operation, and would be invalid and unexecutable scheduling instructions.

[0067] At the same time, the basic feasible domain is constrained and optimized based on several or a specific constraint condition under a particular scenario.

[0068] The constraints include ramp rate constraints, energy constraints, and user service constraints, which will be explained in detail later.

[0069] S300 calculates the corresponding health score based on the communication status and historical performance deviation of each adjustable resource.

[0070] The core of this step is to quantify the operational reliability of resources and generate a health score value within the [0,1] range. The health score is obtained by fusing two types of data: communication quality and performance. First, the original indicators of the two types of data are normalized to the [0,1] range, and then fused and smoothly updated. This solves the core pain point of virtual power plants that "resource reliability cannot be quantified and cannot be incorporated into scheduling decisions" for a long time.

[0071] The health score is directly linked to the resource's callability and scheduling priority, creating a positive incentive where better performance and more stable communication result in a higher health score, which in turn leads to stronger callability and higher returns.

[0072] S400 performs a reliable derating of the basic feasible region based on several health scores to obtain a committed feasible region that includes the second up and down power boundaries.

[0073] By using health scores, the basic adjustable capacity corresponding to the basic feasible region is further multiplicatively reduced to update the upward and downward adjustment boundaries to the promised upward and downward adjustment capabilities. This narrows the resource allocation boundary from the "theoretically adjustable range that can be physically achieved" to the "reliable adjustable range that can be stably fulfilled in practice without default," thus obtaining the promised feasible region. .

[0074] Committed Feasible Domain It is the basic feasible domain A strict subset of the resource, which mandates that all scheduling instructions must match the current reliability level of the resource, prohibits issuing instructions to low-health resources that exceed their reliability capabilities, significantly reduces the deviation rate of scheduling execution, and ensures the accuracy of aggregated responses.

[0075] S500 calculates the adaptive margin based on the health score and prediction error level of all adjustable resources, and reduces the feasible region based on the adaptive margin to obtain the executable feasible region including the third up and down power boundary.

[0076] The core of this step is to dynamically adjust the scheduling safety margin based on aggregated health and prediction error, and further converge the feasible domain based on the safety margin, so that the reliability of each adjustable resource can be further "reserve" for risk mitigation.

[0077] By using adaptive thresholds, more safety margins are reserved to mitigate risks and reduce the probability of default when overall system reliability declines and power prediction errors increase. When the system is running stably and with high reliability, the margin is automatically reduced to release regulation capacity and improve profitability. This solves the problem of either being overly conservative or experiencing uncontrolled risks under static fixed margins.

[0078] Adaptive thresholds are used to define the feasible region. After reduction, the obtained confidence boundary is It is a domain that can be committed to and is feasible. Further optimization of subsets.

[0079] S600 constructs rolling scheduling optimization based on external scheduling instructions and feasible execution domains to obtain and execute scheduling schemes for each adjustable resource.

[0080] The multi-resource regulation problem of virtual power plants is modeled as a convex quadratic programming (QP) problem, which makes the scheduling solution unique and computable under the constraints of health reliability derating and system-level margin.

[0081] By setting an objective function, the system balances aggregated power tracking accuracy with resource usage costs. Weight parameters are configured to flexibly adapt to different operational objectives. A rolling scheduling optimization problem is used to solve for the optimal scheduling scheme, automatically prioritizing core adjustment tasks for high-health, low-cost resources and automatically marginalizing low-health, high-cost resources. This ensures aggregated performance while minimizing scheduling operating costs and equipment wear and tear, and extending resource lifespan.

[0082] In other embodiments, resource capability and constraint modeling is performed based on resource information to obtain a basic feasible region including a first up-down power boundary, including the following steps:

[0083] S210 generates the first up and down power boundaries based on the maximum up-adjustment capability and the maximum down-adjustment capability combined with the online status.

[0084] For each adjustable resource i, the maximum upscaling capacity is collected in real time at time k. and maximum downsizing capability The two parameters serve as the basic boundary parameters, both of which are greater than 0 and are in kW.

[0085] Based on the aforementioned maximum adjustment capacity, multiplicative calculations are performed in conjunction with the online state, and in the offline state... 0. Both the upper and lower power limits are forcibly set to 0, achieving automatic locking of offline resources. For the upper component... Apply the inequality constraint "0 ≤ up-adjustment component ≤ maximum up-adjustment capacity after online linkage" to the down-adjustment component. By applying the inequality constraint "0 ≤ down-adjustment component ≤ maximum down-adjustment capacity after online linkage", the resource base power boundary is defined.

[0086] but:

[0087] .

[0088] S220 generates a ramp rate constraint based on the ramp rate. When the ramp rate is greater than 0, the absolute difference in the resource adjustment amount between adjacent time points is less than the ramp rate.

[0089] First, for each adjustable resource i, obtain its pre-configured ramp rate within a unit scheduling step. ( (kW / step). Where resources without ramp limitations can be assigned a value... This constraint check will be automatically skipped.

[0090] Real-time calculation of resource adjustment amount at time k Resource adjustment amount compared to the previous moment The absolute difference, which represents the rate of change of the resource adjustment quantity.

[0091] Under the climbing constraint, the absolute difference calculated above is compared with the climbing rate. Perform comparison, apply Inequality constraints ensure that the rate of change of resource adjustment does not exceed the physical limits of the equipment, thus avoiding equipment damage or adjustment failure.

[0092] S230 generates energy constraints based on energy information, so that when the resource adjustment amount is increased, the corresponding real-time energy decreases and the real-time energy constraint is within the preset energy capacity boundary.

[0093] In some energy storage / mobile energy load scenarios, energy constraints are also included. The energy information contains pre-configured energy capacity boundaries. and And collect the energy state of k at the current moment in real time. (Unit: kWh, corresponding to the State of Charge (SOC) of the energy storage).

[0094] According to the scheduling step size The unit is used to determine the integral conversion factor for power to energy. This converts the kW-level power regulation into the kWh-level energy change rate, thus unifying the stiffness.

[0095] According to the upward adjustment The unified rule for reducing energy through discharging or generating power is determined by the resource adjustment amount at the current time k. Recursively calculate the energy state at the next time step k+1. The recursive formula is: .

[0096] Apply energy safety boundary constraints to the real-time energy state at the current time t. Apply The inequality constraints ensure that the charging and discharging of energy storage and the adjustment of movable energy loads do not exceed the energy range for safe operation of the equipment, thus avoiding risks such as overcharging and over-discharging.

[0097] At the same time, if it is not an energy storage resource, the energy state need not be introduced. This could result in an extremely wide energy capacity boundary.

[0098] S240 generates business constraints based on user business information, ensuring that the function value can be calculated after substituting the given resource adjustment amount, real-time energy and external business volume into the explicit constraint expression and that it is not greater than 0.

[0099] First, the scenario is constrained and parameters are analyzed. For each resource i, its network connection restrictions, operational requirements, business rules, and other personalized constraints are analyzed. The relevant external measurable / known quantities are identified and defined as external business quantities. External business volume This includes information such as photovoltaic available power forecasts, grid connection point restrictions, EV off-site timing, and energy demand.

[0100] All personalized constraints are uniformly converted into explicit inequality constraint functions. ( ) The form ensures that the constraint function can be directly calculated and the constraint satisfaction can be determined after given the adjustment amount, state amount, and external amount.

[0101] Adjust the current amount of candidate resources Adjustment amount at the previous moment Real-time energy status External context traffic Substitute the constraint function and calculate the function value.

[0102] Finally, the constraint compliance is determined by verifying whether the calculation results of all constraint functions meet the requirement of not being greater than 0. Only when all constraints are met is the candidate adjustment quantity determined to be compliant and included in the adjustable capability range.

[0103] Specifically, the constraint function can be expressed in inequality form as follows:

[0104] .

[0105] The operational characteristics of different types of distributed resources, such as photovoltaics, energy storage, EVs, and controllable loads, are uniformly converted into explicit and computable inequality constraints, breaking down industry barriers that prevent different types of resources from being included in a unified scheduling framework, and realizing standardized aggregation modeling of multi-source heterogeneous resources.

[0106] By defining four categories of constraints—power boundary, ramp-up, SOC, and user services—the maximum adjustable range for resource physical safety, grid connection compliance, and user wishes is precisely defined, ensuring that all scheduling instructions do not exceed equipment operating limits or violate grid connection rules and user agreements.

[0107] The following examples illustrate how to form a basic feasible region:

[0108] A: Photovoltaic (PV) resource constraints:

[0109] Let the upper limit of available output of photovoltaic resources at time k be... ,in, The net resource injection power is defined as obtained from irradiance prediction or actual measurement:

[0110] .

[0111] in, Characterized as baseline or planned power, the photovoltaic output constraint can be expressed as:

[0112]

[0113] If photovoltaic ramping limitations (grid connection or inverter control requirements) are considered, a given ramping upper limit is given. (kW / step) can be written as:

[0114] .

[0115] If you need to consider the apparent power limitation of the inverter and simultaneously regulate reactive power... Given capacity Then it can be expressed as:

[0116] .

[0117] B: Energy storage resource constraints (safe SOC / power / lifetime, etc.):

[0118] Energy storage resources introduce energy state Energy capacity boundary And use up / down decomposition ,but:

[0119] ,

[0120] .

[0121] Energy storage charge / discharge power boundary (given maximum discharge power) With maximum charging power ), can be written as:

[0122] ,

[0123] .

[0124] If further limits are needed on daily throughput / cycle intensity related to battery life, cumulative throughput energy can be defined. (Unit: kWh), its calculation method is as follows:

[0125] .

[0126] Set the throughput limit to ,but:

[0127] .

[0128] C: EV (charging station / vehicle fleet) resource constraints:

[0129] EV resources at all times Connection status This indicates that 1 represents a connected window and 0 represents an unconnected window. Therefore, to force the window to be unresponsive outside the connection window, it can be written as:

[0130]

[0131] in This is the upper limit of the maximum discharge power. This represents the maximum charging power limit. Furthermore, let the EV departure time be... The minimum energy requirement for leaving the station is To ensure that at any given time... All can meet the "meeting the standard before leaving the station" requirement, and the remaining steps are taken into account. And construct reachability constraints: In this step, update to Afterwards, even if it continues to be charged at the maximum charging power, it should still meet the requirements:

[0132]

[0133] Equivalently written as computable inequalities:

[0134] .

[0135] in Calculated according to the aforementioned energy renewal equation For charging efficiency.

[0136] D: Well network point / distribution network capacity constraints:

[0137] To meet the power limit of the point of connection (PCC), the maximum external power limit of the PCC is set as follows: The maximum power receiving limit is Then it can be written as:

[0138] ,

[0139] .

[0140] in This refers to the net injection power of each resource under a unified standard.

[0141] Through the methods described in the examples above, user / business constraints can be explicitly represented as a set of computable inequalities, which can be directly calculated given state variables, external variables, and candidate adjustment variables. The value is determined and the constraint is satisfied, thus enabling seamless coupling with the health reliability degradation and gradual allocation scheduling process.

[0142] Suppose a certain resource at time... :

[0143] First, the power boundary is given: ;

[0144] Apply hill climb restrictions: Previous moment ,and ,but

[0145] Further tighten SOC restrictions: for example, only allow (No more discharge is possible);

[0146] Further tightening of business constraints: for example, grid connection / planning requirements limiting it to a maximum of 40, thus achieving...

[0147] So the final basic feasible region It's the section where all these restrictions are "commonly permitted":

[0148] .

[0149] S250 integrates the resource adjustment amount after satisfying the first upper and lower power adjustment boundaries and all constraints into the basic feasible region.

[0150] In other embodiments, a corresponding health score is calculated based on the communication status and historical performance deviations of each adjustable resource, including the following steps:

[0151] S310 acquires online rate, command receipt success rate, packet loss rate, and round-trip latency, performs scale unification and normalization, and calculates the communication score at each time point by combining the communication weight set.

[0152] Input measurable communication metrics and collect these metrics in real time. For each resource i at time k within the collection and statistics window, the core communication metrics specifically include:

[0153] Online rate / Heartbeat success rate: (For example, recently) (Heartbeat success rate);

[0154] Command Response Success Rate: ;

[0155] Loss rate: ;

[0156] Round trip delay: (ms).

[0157] Meanwhile, regarding packet loss rate Round-trip delay The two values ​​belong to the category of negative indicators where "the larger the value, the worse the reliability," and therefore require positive normalization. This is achieved by mapping both indicators to an exponential decay function. Positive ratings within the interval, where:

[0158] ,

[0159] .

[0160] in For scale parameters (e.g., you can fill in the following): (etc.), the smaller the value, the more sensitive the score is to changes in the indicator. Through the above conversion, the above indicators are also uniformly converted into the expression "the larger the value, the higher the reliability".

[0161] Pre-configure the weights corresponding to the above communication metrics, respectively. , , , Furthermore, the weights are forced to sum to 1, and the weight values ​​represent the system's emphasis on the corresponding communication metrics.

[0162] The above-mentioned normalized communication indicators are then weighted and summed according to preset weights to obtain an initial communication score. Further processing is then performed... The function forces the initial communication score to be within the range [0,1]. Any value exceeding 1 is truncated to 1, and any value below 0 is truncated to 0, resulting in the final communication score. The specific calculation formula is as follows:

[0163] .

[0164] S320: Obtain the deviation amplitude of the target adjustment amount and the actual adjustment amount of each adjustable resource at the same time, divide it by the reference power base to calculate the relative error, obtain the response time of each adjustable resource to the command, normalize the relative error and response time through the exponential decay normalization function to obtain the error sub-score and time sub-score, obtain the alarm index of the adjustable resource at each time and normalize it to generate the alarm score.

[0165] First, for each resource i at time k, collect the target adjustment amount issued by the scheduling system. Adjustment amount of actual response .

[0166] Extract the maximum values ​​of the current maximum up-adjustment capacity and maximum down-adjustment capacity of the resource, and use them as the reference power benchmark for error normalization. It is used to eliminate the differences in dimensions and scales between resources of different capacities. Simultaneously, it sets a small positive number. To avoid division by zero errors when the reference power is 0, the above parameters are used to quantify the performance of resource i in fulfilling and controlling its discrete time k. The relative error between the command and the actual measurement is defined to eliminate the influence of the absolute value of the power on the error judgment. The specific formula is as follows:

[0167] .

[0168] in, Indicates the scheduling system at time Distribute resources The target adjustment amount (in kW); Representing resources At any moment The actual adjustment amount (in kW, which can be obtained from the difference between the measured power and the baseline power); This represents absolute value operations and is used to characterize the magnitude of the deviation. The reference power standard (in kW) representing error normalization is used to eliminate dimensional and scale differences between different capacity resources, and is preferably selected.

[0169]

[0170] in and Representing resources At any moment Maximum upward and maximum downward adjustment capabilities (in kW); To avoid A tiny positive number (unit: kW) appears when divided by zero.

[0171] To address relative errors, response time is introduced synchronously. (Unit: seconds) is defined as the time it takes for a resource to reach a steady state, 90% arrival time, or other equivalent time measure after receiving an instruction, and is used to characterize the speed of resource response.

[0172] Meanwhile, for the two negative indicators, relative error and response time, an exponential decay function is also used to map them to positive scores within the interval [0,1], where:

[0173] ,

[0174] .

[0175] in, Represented by natural constant An exponential function with base 0; Indicates relative error The error score obtained from the mapping Indicated by response time The response speed score obtained from the mapping; This is a dimensionless decay scale parameter for the error score, used to adjust the score's sensitivity to changes in error. The decay scale parameter (in seconds) for the response score is used to adjust the sensitivity of the score to changes in response time. .

[0176] Furthermore, anomaly / alarm indicators are introduced:

[0177] .

[0178] It is used to characterize resource i at time i. The alarm status or abnormality level can preferably be set to 1 when there is no alarm and 0 when there is an alarm, or mapped according to the alarm level. The score within the interval.

[0179] S330 calculates the performance score based on the error sub-score, time sub-score, and alarm score combined with the corresponding performance weight set.

[0180] Finally, obtain the performance weights corresponding to the three pre-configured metrics: relative error, response time, and alarms. , , The weights are forcibly summed to 1, and the three normalized indicators are weighted and summed according to their weights to obtain the initial performance score.

[0181] At the same time, the clip function is used to force the initial performance score to be constrained within the interval [0,1] to obtain the final performance score. The specific calculation formula is as follows:

[0182] .

[0183] S340 uses a weighted fusion of communication score and performance score, and employs an exponentially weighted moving average for smooth updates to obtain a health score.

[0184] First, configure the fusion weights, and then assign the configured fusion weights, which correspond to communication scores and performance scores respectively. , Furthermore, the sum of the two weights is forced to be 1, and the weight size represents the system's emphasis on communication reliability and contract fulfillment reliability.

[0185] The communication score and performance score are weighted and summed according to the above-mentioned fusion weights. The calculation result is constrained to the interval [0,1] by the clip function to obtain the real-time comprehensive score at the current moment. This score reflects the instantaneous reliability of the resource at the current moment. The calculation formula is as follows:

[0186] .

[0187] Configure the smoothness coefficient for health updates Generally take The closer the coefficient is to 1, the more stable the health update and the stronger the resistance to instantaneous fluctuations. The closer the coefficient is to 0, the more sensitive the health is to changes in real-time scores.

[0188] By using the first-order exponential smoothing formula, combined with the health score from the previous time step, And the current real-time comprehensive score Calculate the final health score at the current moment. Initial health score The default value is 1. The specific formula is:

[0189] .

[0190] Resources with higher health levels have commitment capabilities closer to their physical adjustability limits, allowing them to release more regulatory capacity to participate in the market and generate revenue. Conversely, resources with lower health levels have automatically reduced commitment capabilities, limiting their scale of participation in scheduling. This truly achieves "automatic conservative control of defaults when risks are high, and releasing capacity to increase revenue when risks are low," resolving the revenue loss problem caused by static conservative commitments.

[0191] In other embodiments, a credible derating of the basic feasible region is performed based on several health scores to obtain a committable feasible region including a second up-and-down power boundary, including the following steps:

[0192] S410, the calculated health score is used as a confidence coefficient, and the first up and down power boundary is multiplied by the confidence coefficient to calculate the second up and down power boundary.

[0193] Score the health of each resource i at time k. As a reliability coefficient, the closer the health score is to 1, the higher the resource reliability and the closer the reliability coefficient is to 1.

[0194] The basic maximum upward / downward adjustment capability of a resource is multiplicatively derating using a reliability coefficient. Specifically, the first upward / downward adjustment power boundary is multiplied by the reliability coefficient to calculate the second committable upward / downward adjustment power boundary. This ensures that the lower the reliability of the resource, the smaller the committable adjustment capability. The specific formula is as follows:

[0195] ,

[0196] .

[0197] Using the reduced, committable upward and downward adjustment capacity as the new boundary for adjustment of resource volume. Apply the corresponding scheduling constraints, specifically:

[0198] .

[0199] The range of adjustment that simultaneously satisfies both the first and second power adjustment boundaries is defined as the feasible region that can be committed to. This set is the core feasible region for subsequent scheduling optimization, ensuring that scheduling instructions are issued based on the reliable capabilities of resources.

[0200] In other embodiments, an adaptive margin is calculated based on the health scores and prediction error levels corresponding to all adjustable resources, including the following steps:

[0201] S510 generates weighted weights based on the maximum adjustment capacity of each adjustable resource, and calculates the aggregate health score by combining the health score with the weighted weights and performing a weighted average.

[0202] First, the weighting weights are determined, and the maximum adjustment capacity of each resource i at time k is obtained. As a weighted average, resources with larger capacities have a greater impact on the overall aggregate health, which better reflects the actual operating characteristics of virtual power plants. The specific formula is as follows: .

[0203] The health scores of all resources are summed using the weighted averages generated above, resulting in a weighted total health score. This weighted total health score is then divided by the sum of all weights, with a small positive factor added. To avoid division by zero errors, the overall aggregate health of the virtual power plant is ultimately obtained. Its value ranges from [0,1], reflecting the overall reliability level of the entire virtual power plant. The specific formula is:

[0204] .

[0205] The S520 collects the measured power response data from the previous moment and the external dispatch command from the previous moment, and combines them with the preset aggregated reference power benchmark to calculate the aggregated tracking error.

[0206] Collect the aggregate measured power response at the previous time k-1 External dispatch instructions The aggregate tracking relative error of the previous time step is calculated using the following formula.

[0207] ,

[0208] in As a aggregated reference power standard, its specific value can be taken as... Or a fixed benchmark, To prevent tiny positive numbers from being divided by zero.

[0209] Specifically, It is used to quantify the actual performance and default degree of the virtual power plant at the previous moment; the larger the value, the more serious the performance deviation. The output of the virtual power plant as actually perceived by the grid side represents the true result of the virtual power plant's performance. The dispatch and tracking instructions issued to external parties or the target values ​​promised to external parties are the legal obligations of virtual power plants and the baseline for judging whether there is a breach of contract.

[0210] S530 obtains the smoothing parameters and combines them with the aggregated tracking error. It then calculates the prediction error variance at the current time using the exponentially weighted moving average method and takes the square root of the prediction error variance to calculate the prediction error level.

[0211] Pre-configure the smoothing coefficient of EWMA The closer the coefficient is to 1, the stronger the dependence of the subsequent prediction variance on historical errors and the more stable the change. The closer the coefficient is to 0, the more sensitive the response to the latest error.

[0212] The prediction error variance at the current time is calculated recursively using EWMA (Exponentially Weighted Moving Average), also known as exponential smoothing.

[0213] .

[0214] Among them, the variance of the prediction error at the initial time .

[0215] The variance of prediction error is a core indicator that represents the "amplitude of data fluctuation and the magnitude of uncertainty". The larger the value, the more drastic the fluctuation of the virtual power plant's future tracking error and the higher the uncertainty of default.

[0216] For prediction error variance Perform a square root calculation to obtain the prediction error level at the current time. This reflects the uncertainty of virtual power plant power tracking, and the specific formula is as follows:

[0217] .

[0218] By using weighted recursion, the fluctuation risk of virtual power plant tracking error during future dispatch cycles can be accurately predicted, providing a quantitative basis for setting subsequent safety margins.

[0219] S540, the aggregated health level and the prediction error level are multiplied by the equilibrium coefficient and summed to obtain the adaptive margin.

[0220] Pre-configure upper and lower limits of margin Limit the range of margin fluctuations to avoid being overly conservative or overly aggressive. The margin range is generally within... Depending on the business, the higher the value, the more conservative the system and the lower the probability of default; conversely, the higher the value, the more aggressive the system and the higher the potential for profit.

[0221] Simultaneously configure the tradeoff coefficient This is used to balance the weights of the prediction error level and the aggregate health on the margin.

[0222] The prediction error driving term is obtained by multiplicatively calculating the aggregate health score and the prediction error level with the equilibrium coefficient, respectively. and health-driven items The sum of the two terms represents the initial margin calculation cost, which is also based on the clip function to force the initial margin to be constrained to the upper and lower limits of the margin. , Between ], the adaptive margin at the current time is obtained, and its specific calculation formula is as follows:

[0223] .

[0224] In other embodiments, the following steps are also included:

[0225] S541: Obtain the current operating scenario. If the current operating scenario is short-term fluctuation, increase the equilibrium coefficient; if the current operating scenario is long-term reliability, decrease the equilibrium coefficient.

[0226] Depending on the scenario, the size of the tradeoff coefficient can be configured accordingly. The larger the balance coefficient, the more the change in adaptive margin depends on the change in the prediction error level, making it more sensitive to short-term fluctuations and responding faster. The smaller the balance coefficient, the more the change in adaptive margin depends on the change in aggregate health, resulting in a more stable change.

[0227] Therefore, when the trade-off coefficient is too large, it is suitable when the error index can well reflect the risk and the system frequently experiences short-term instability (such as rapid changes in PV caused by cloud shadows, large fluctuations in EV group charging, and frequent commands). However, the potential risk is that if the prediction error level is too high... High noise levels or unstable estimates can cause margin fluctuations, making scheduling more "tight and loose," which may affect profitability and stability.

[0228] When the trade-off factor is small, it is suitable for situations where health (communication + performance history) is trusted more as a source of risk, or when the error measurement itself is extremely unstable / easily contaminated by external noise. However, its potential risk is that the response to sudden increases in error will be relatively slow (for example, when the cloud shadow suddenly goes offline at a certain time, causing the error to rise, but the health has not yet had time to decrease).

[0229] In other embodiments, reducing the committable feasible region based on adaptive margin to obtain an executable feasible region including a third up-and-down power boundary includes the following steps:

[0230] The S550 calculates the safety margin factor through adaptive margin calculation, and multiplies the second upper and lower power adjustment boundaries by the safety margin factor to calculate the third upper and lower power adjustment boundaries.

[0231] The safety margin factor is defined by calculating the adaptive margin, specifically by using 1- What was obtained.

[0232] After obtaining the safety margin coefficient, the confidence boundary after further margin reduction, i.e., the third power adjustment boundary, is obtained by multiplying it with the second power adjustment boundary. The specific formula is as follows:

[0233] ,

[0234] The above range corresponds to the feasible execution domain. .

[0235] Subsequently, when solving the objective function, the decision variables corresponding to the resource adjustment amounts in each time period within the rolling window will be used. Need to meet The boundary constraints are determined simultaneously by considering the aforementioned constraints such as ramping, energy, and users.

[0236] It actually implicitly uses several preceding parameters. First, it uses health to make a reliable reduction, and then uses margin m to leave a system-level safety margin. When the overall health decreases or the prediction error increases, it automatically increases the margin to reduce the probability of default; when the health is stable, it decreases the margin to increase returns.

[0237] The subsequent objective function is an optimization in form. However, its optimization space has been gradually constrained and shaped by the aforementioned resource information, including time discretization, physical and business constraints, health assessment, credibility reduction, and risk margin. Therefore, the subsequent objective function does not determine the final result; the final decision is based on the objective function multiplied by the feasible region.

[0238] In other embodiments, a rolling scheduling optimization is constructed based on external scheduling instructions and the feasible execution domain to obtain and execute a scheduling scheme for each adjustable resource, including the following steps:

[0239] S610, configure the window length and the time period that needs to be optimized for each scheduling moment.

[0240] Pre-set the window length H for rolling optimization, and determine the future time period that needs to be optimized at each scheduling time k. The quadratic programming (QP) problem will be solved within each rolling length.

[0241] S620 uses the resource adjustment amount of each adjustable resource in each time period within the rolling window as the decision variable.

[0242] Adjustment amount of each resource i in each time period within the scrolling window As a decision variable in the optimization solution.

[0243] S630, construct the objective function of the quadratic programming, with the goal of minimizing the overall cost.

[0244] Construct a quadratic programming objective function to minimize the overall system cost. The function consists of two terms: the first term is a quadratic term that aggregates the total error. The first term represents the priority weight; the higher the weight, the higher the system's requirement for tracking accuracy. The second term is a linear term representing the resource usage cost. Let be the cost / preference coefficient for resource i. The larger the coefficient, the more the scheduler tends to avoid calling that resource. The value can be 0 or adjusted according to the quoted price.

[0245] The final objective function is as follows:

[0246] .

[0247] Loading constraints, such as aggregation tracking constraints, can be either hard constraints where the aggregation adjustment is strictly equal to the external scheduling instruction, or soft constraints implemented through the quadratic term of the objective function.

[0248] S640, within the feasible region, find the optimal decision solution that minimizes the objective function to obtain the optimal adjustment sequence for each time period within the rolling window.

[0249] S650 executes only the optimal adjustment value corresponding to the first time period in the optimal adjustment value sequence and distributes it to the corresponding adjustable resources.

[0250] The quadratic programming solver is invoked to find the optimal solution of the decision variables corresponding to minimizing the objective function within the feasible region formed by the loaded constraints, thereby obtaining the optimal adjustment sequence of each resource i for each time period j within the rolling window.

[0251] At the same time, only the optimal adjustment command for the first time period within the rolling window (i.e., the current time k) is executed and sent to the corresponding resource. The optimization results for subsequent time periods within the window are used as a reference and are not executed.

[0252] Finally, during the rolling iteration, when entering the next scheduling time k+1, the above steps are repeated. The optimization problem is reconstructed and solved based on the latest state data to achieve closed-loop rolling scheduling.

[0253] Detailed explanation:

[0254] The objective function ultimately solves not an abstract economic scheduling problem, but rather the optimal trade-off between aggregation tracking error and resource usage cost within a feasible region where health is reliable and risk is controllable. In essence, the objective function defines a "system regret value / loss value".

[0255] Regarding the aggregate tracking error corresponding to the first item, the larger the corresponding deviation, the greater the corresponding loss value, and the impact is further amplified by the square of the deviation. This item is not concerned with a single resource, but rather with whether all resources aggregated together have kept up with the scheduling instructions required by the system.

[0256] Priority weight The larger the value, the higher the priority of tracking; the system will make the most effort to meet the power requirements of the scheduling command. The hourly rating indicates that a certain level of tracking error is acceptable, and the system prioritizes cost and resource protection. This priority weight determines whether it is a grid-friendly virtual power plant or a cost-conservative virtual power plant.

[0257] Regarding the usage cost of the second item, it does not care whether the tracking was successful, but rather which resources and how many resources were used to complete the scheduling task.

[0258] Use coefficient It represents the cost of using resources. The larger the value, the less willing the scheduler is to use that resource. It can represent various information objects, such as actual quotes / compensation costs, battery life depletion weights, penalties for devices that do not want to be frequently scheduled, implicit penalties for communication instability and low health, user preferences, etc.

[0259] Combining the above two points, and At each moment j, the system makes a trade-off:

[0260] "In order to get even closer" Is it worthwhile to use another resource?

[0261] In other words, it's a comparison between reducing the squared tracking error or increasing resource usage costs.

[0262] The behavior of min in the objective function during rolling scheduling is as follows: at each time step k, the system constructs an optimization problem (QP), calls the solver to make a decision using min within the constrained space, and outputs a complete segment. Trajectory, and only one execution. Then, repeat the above process once more at the next moment.

[0263] Therefore, `min` is a recurring online decision-making process: given physical constraints, health reliability constraints, and system-level margin constraints, it globally compares multi-resource, multi-time-period adjustment schemes and selects the adjustment sequence that minimizes the system's overall cost function (objective function), thereby achieving the optimal trade-off between external tracking performance and resource usage costs. It doesn't simply calculate a numerical value, but rather selects the least regrettable scheme from all feasible scheduling options under the current understanding and risk conditions.

[0264] As explained above, the minimization operation (min) in this scheme is not a simple mathematical calculation, but rather a global comparison and automatic selection of multiple resource adjustment schemes within the feasible region jointly shaped by health reliability degradation and system-level margin. This mechanism can explicitly reduce the impact of unreliable resources on system operation while ensuring aggregate fulfillment capabilities, thereby achieving a balance between reliability, economy, and feasibility.

[0265] The following examples illustrate QP decision-making:

[0266] Scenario Example 1 (Simple Scenario):

[0267] I. Problem Setting:

[0268] In this scenario, if only one time point is set (H=1), then the current time point j=k, and the corresponding external scheduling instruction is: .

[0269] There are three adjustable resources. The adjustable range of resource 1 is... Costs in the range of 0-10 A value of 1 corresponds to the meaning of cheap and easy to use, and the adjustable range of resource 2. Costs in the range of 0-10 The value is 2, which corresponds to the general meaning, and the adjustable range of resource 3. Costs in the range of 0-10 A score of 5 indicates that the item is very expensive and the user does not wish to use it.

[0270] At the same time, the resource adjustment amount is set to only be allowed to be increased, that is... .

[0271] Set tracking priority The objective function is then:

[0272] .

[0273] The min task is among all the valid tasks. Find one in the final The smallest.

[0274] II. The decision-making process includes:

[0275] Option A: Use only the cheapest resource 1, then = (10, 0, 0).

[0276] Tracking error is Resource costs are The final cost was .

[0277] Option B: Average resource usage is 1 or 2, then = (5, 5, 0).

[0278] The tracking error is 0, and the resource cost is The final cost was Therefore, it can be concluded that this solution is worse than solution A.

[0279] Option C: All resources are allocated equally, then =(3.33,3.33,3.33).

[0280] The tracking error is 0, and the resource cost is The final cost was This plan is even worse.

[0281] Option D: Do not use resource 3, but reduce the number of scheduling instructions slightly. = (8, 0, 0).

[0282] Tracking error is Resource costs are The final cost was .

[0283] Option E: Use a slightly more expensive resource (3) to reduce the error. = (8, 0, 2).

[0284] Then the tracking error is Resource costs are The final proxy result is The proposed solution is still worse than Solution A.

[0285] In summary, among all the above options, the result of using the min option to make a decision is: .

[0286] It satisfies the system objective, namely, the tracking error is 0; secondly, it reflects resource preference, with the most expensive resource 3 having a value of 0, and the second most expensive resource 2 also having a value of 0.

[0287] three, Change:

[0288] if If this is changed, the subsequent decisions made by min will also change, such as... When =1, it indicates that the tracking error is not so important; in this case, the tracking error corresponding to scheme D becomes... The resource cost is 8, and the final And Option A The value remains 10, so we still choose option A, but the actual difference becomes smaller.

[0289] Scenario Example 2 (Adding Health / Margin Constraints):

[0290] 1. Retain the original settings of the previous example and introduce an additional health level h.

[0291] Assume the current health levels are as follows: Resource 1 corresponds to 1.0 (very reliable), Resource 2 corresponds to 0.5 (average), and Resource 3 corresponds to 0.2 (very unstable).

[0292] The capabilities after the credibility reduction are as follows: Resource 1 For 10, resource 2 For 5, resources 3 The value is 2.

[0293] Introducing a system-level margin m, set to 0.2, the final usable boundary is... Therefore, the final upper limit of resource 1 is... The final limit for resource 2 is The final limit for resource 3 is .

[0294] At this point, it should be noted that the optimal solution (10,0,0) in the previous example is invalid because it exceeds the maximum limit of resource 1, and the resources need to be adjusted accordingly.

[0295] In this example, the objective function remains the same:

[0296] .

[0297] II. The decision-making process includes:

[0298] New Option A: Use only resource 1, at which point the corresponding .

[0299] The polymerization power is 8, and the tracking error is... The resource cost is 8, and the final cost is J = 40 + 8 = 48.

[0300] Option B: Use up all resources 1 + replenish with resources 2. .

[0301] The aggregation power is 10, the tracking error is 0, and the resource cost is... The final cost is J = 12 + 0 = 12.

[0302] Option C: Utilize both Resources 1 and Resource 2 to their full potential. .

[0303] The polymerization power is 12, the tracking error is 40, the cost is 16, and the final cost result is J=56.

[0304] Option D: Use resource 3 simultaneously. .

[0305] The polymerization power was 10.8, and the tracking error was [missing information]. The cost is The final cost is J=22.4.

[0306] III. The final choice of min:

[0307] The optimal solution selected based on J from the above-mentioned schemes is... Because the J value is the smallest under this scheme.

[0308] IV. Comparison with Example 1 when there is no health / margin:

[0309] The optimal solution without health status is (10,0,0), and the optimal solution with health status and margin is (8,2,0). It can be observed that with health status and margin, the system will no longer exploit resource 1, and will hardly use the unhealthy resource 3. The system will automatically transfer the burden to the "relatively reliable" resource 2, and the aggregated tracking results are still accurate.

[0310] Health and margin do not change the form of the scheduling objective function. Instead, by dynamically shrinking the feasible domain of decision variables, the minimization operation is carried out within a reliable and conservative capability space, thereby prompting the optimal solution to automatically avoid unreliable resources and reasonably distribute the adjustment tasks.

[0311] In other embodiments, when resource disconnection, latency spikes, or consecutive unanswered requests are detected, a rapid decline in health and immediate capacity reduction are triggered, and external commitments are reallocated or downgraded.

[0312] Specifically, during the rolling scheduling process, when the operating or communication status of any adjustable resource meets the preset anomaly triggering conditions, a corresponding anomaly triggering flag and anomaly confidence level are generated. If the anomaly triggering flag is valid, the health update rule for that resource is switched, and the resource's health is multiplicatively updated based on the anomaly confidence level using a fast decay parameter. This shortens the effective timescale of the resource's health after an anomaly occurs and correspondingly compresses its reliable adjustment capability for scheduling. Specifically, this also includes:

[0313] S700: When any adjustable resource is detected to meet the preset abnormal triggering conditions, its health score is updated using the fast decay parameter, and the calculation of the second up and down power boundary, the calculation of the adaptive margin, and the solution of the rolling scheduling optimization are re-executed.

[0314] First, pre-configure the trigger thresholds for various abnormal indicators, including the online rate threshold. Response rate threshold Delay threshold Packet loss rate threshold Performance error threshold For each resource i at time k, several core metrics are checked in real time to see if they have abnormally triggered the trigger thresholds. These metrics include:

[0315] Low online rate: ;

[0316] Low response rate: ;

[0317] High latency: ;

[0318] The bag was dropped too high: ;

[0319] Excessive performance error: .

[0320] At the same time, configure the number of consecutive steps L for abnormal triggering to avoid false triggering caused by a single instantaneous fluctuation. When performing abnormal judgment in the future, you can choose to trigger if any condition is met or if the cumulative number of triggers for any condition exceeds L.

[0321] "Continuous" can be added "Step satisfies": For example, if any of the above conditions are met in the most recent The cumulative number of steps to satisfy Then it will be triggered.

[0322] When the triggering conditions are met, an abnormal trigger flag is generated, marking the resource as an abnormal state and entering the subsequent rapid health decay process. Resources that do not meet the conditions maintain normal health and continue the update process.

[0323] Once the condition is triggered, the pre-configured fast decay parameter is retrieved. and the lower limit of health safety Among them, the rapid decay parameter The value is generally between 0.2 and 0.7, with the specific value determined by the anomaly confidence level. The more severe the anomaly, the higher the confidence level. The smaller the value, the faster the health status decays. This anomaly confidence level represents the comprehensive severity of all anomalies and can be obtained by normalizing and weighting all anomalies.

[0324] Instead of the conventional smooth update formula, a multiplicative fast decay method is used to calculate the current health of abnormal resources, achieving a rapid decrease in health after an anomaly occurs. The specific formula is as follows:

[0325] .

[0326] Simultaneously, based on the updated health status, the aforementioned reliable deregulation formula is used to recalculate the committable upward / downward adjustment capability of the adjustable resource. This enables the immediate contraction of the adjustable capability boundary of abnormal resources, preventing unreliable resources from continuing to undertake adjustment tasks. The specific formula is as follows:

[0327] .

[0328] Meanwhile, the subsequent adaptive margin and rolling scheduling optimization solutions are also recalculated based on the new feasible region.

[0329] The following section details the execution process of the aforementioned health status drop trigger mechanism, using specific time-series events as examples: When resource operating status or communication quality undergoes a sudden change, how does the system perform health status updates, capacity contraction, and scheduling result reconstruction in stages and computationally, thereby avoiding aggregate default risks?

[0330] (I) Initial scheduling state (before the event occurs):

[0331] Assuming at the rolling scheduling time The virtual power plant has completed the scheduling solution and is operating stably. The system status is as follows:

[0332] 1) External scheduling instructions: .

[0333] 2) Resource set: Three types of adjustable resources Their basic maximum regulation capacity is 10kW.

[0334] 3) Health status at the previous moment: .

[0335] 4) Based on the reliability and margin conditions, the optimal scheduling result at the previous time step is: At this point, the system's external power tracking error is zero, and it is in a stable performance state.

[0336] (II) Abnormal Event Detection and Trigger Determination:

[0337] At the next scheduling time The system detected a communication anomaly in resource 2 in real time, which manifested as follows:

[0338] The latest statistics window shows the command execution success rate. Below the preset threshold ;

[0339] At the same time, it was detected that no valid power acknowledgment was received within L consecutive scheduling steps.

[0340] According to the judgment rules, the system determines that resource 2 meets the trigger condition of a sudden drop in health.

[0341] [Phase Result 1]: Resource 2 is marked as "abnormally triggered resource" and enters the rapid health decay process; the status of the remaining resources remains unchanged.

[0342] (III) Rapid Health Decline Update:

[0343] The system no longer uses the conventional EWMA smooth update, but instead applies a fast decay rule to resource 2:

[0344] .

[0345] Assuming to take , Then we have:

[0346] .

[0347] The health of Resource 1 and Resource 3 will either be updated normally or remain at their original values, i.e.: , .

[0348] [Phase Result 2]: The health of resource 2 suddenly dropped from 0.5 to 0.15, and its reliability was significantly reduced.

[0349] (iv) Immediate contraction of credibility:

[0350] Based on the updated health status, the system immediately recalculates the trustworthiness and commitment capability of each resource: , , .

[0351] It is evident that the adjustability of resource 2 has been significantly reduced, leaving only a very small amount of usable space to prevent it from continuing to undertake critical adjustment tasks.

[0352] [Phase Result 3]: The credible adjustment capability of resource 2 has decreased significantly, and it no longer meets the conditions for being the main adjustment resource.

[0353] (v) System-level margin recalculation:

[0354] Due to the decrease in aggregate health, the system recalculates the weighted average health. By combining historical prediction errors, a new system-level margin is obtained:

[0355] .

[0356] For example, the margin was increased from 0.2 to 0.3.

[0357] Accordingly, the final available adjustment limit for each resource has been further reduced.

[0358] [Phase 4]: The system as a whole has entered a more conservative operating state, reserving more safety margin to cope with potential uncertainties.

[0359] (vi) Rolling scheduling refactoring:

[0360] Under the new health, reliability, and margin constraints, the system reconstructs and solves the rolling scheduling optimization problem. Since the availability of resource 2 is significantly reduced, it no longer undertakes the main adjustment task in the new optimal solution.

[0361] After solving, the system obtains the following new scheduling result example:

[0362]

[0363] If the aggregated available capabilities are insufficient to fully satisfy external instructions, the system allows for controllable tracking errors, or further triggers the external commitment degradation mechanism.

[0364] [Phase 5]: The scheduling results are automatically reconstructed, abnormal resources are marginalized, and the system as a whole remains executable and controllable.

[0365] As can be seen from the above examples, the health drop triggering mechanism of this invention can quickly complete health decay, immediate capacity contraction, and scheduling reconstruction in an event-driven manner after a resource anomaly occurs. The entire process requires no manual intervention or hard-coded rules, achieving automatic isolation of abnormal resources and system-level risk suppression, thereby significantly improving the reliability and safety of virtual power plants in complex operating environments.

[0366] In other embodiments, it also includes:

[0367] S800: If the health score is lower than the removal threshold, the online status of the adjustable resource is set to 0 and it is removed.

[0368] Pre-configure resource culling threshold When a resource's health score falls below this threshold, it is deemed completely unreliable and is directly removed. Specifically, the resource is identified online. Set to 0, resource adjustment amount Set it to 0 to completely remove the resource from the scheduling resource pool and prevent it from participating in subsequent scheduling optimizations.

[0369] S810, after being removed, recalculate the maximum upward and downward adjustment capabilities corresponding to the virtual power plant.

[0370] Based on the updated resource health and the removed resource pools, the maximum upscaling and downscaling availability of the virtual power plant as a whole is recalculated.

[0371] S820: If the scheduling scheme cannot meet the external scheduling instructions, the external commitment is downgraded to generate new external scheduling instructions to limit them to the range of aggregated available capabilities.

[0372] Meanwhile, if the aggregated available capabilities cannot cover the current external scheduling instructions, the clip function is used to restrict the instructions to the range of aggregated available capabilities to perform commitment degradation and generate new external scheduling instructions. The specific calculation formula is as follows:

[0373] ,

[0374] in,

[0375] .

[0376] This transforms "insufficient system capacity" into an executable degradation commitment.

[0377] The implementation principle is as follows:

[0378] This solution addresses the core pain points of virtual power plants, such as over-commitment and default, excessive conservatism leading to lost revenue, and the inability to quantify and manage reliability. It constructs a complete technical system for health-driven assessment of committable capabilities and adaptive scheduling.

[0379] Technically, the system first establishes a set of heterogeneous resource-based adjustable capabilities covering power, ramping, energy, and service constraints through standardized discrete modeling. Then, it integrates communication quality and historical performance to quantify resource health and completes the mapping from "theoretically adjustable capabilities" to "capable of fulfilling commitments" through reliable derating. Subsequently, based on aggregated health and tracking prediction errors, it dynamically calculates adaptive scheduling margins and constructs an online solvable convex quadratic programming rolling scheduling model, along with a closed-loop anomaly handling mechanism triggered by events.

[0380] In terms of effectiveness, this solution balances operational reliability and economy from the root, significantly reducing default risk while maximizing dispatch benefits; it achieves standardized aggregation and refined management of distributed resources, and can automatically reconstruct the dispatch scheme in abnormal scenarios, significantly improving operational robustness; the entire process is computable and easy to implement in engineering, adapting to the real-time dispatch requirements of the power market, and effectively improving the grid friendliness and market competitiveness of virtual power plants.

[0381] It should be understood that although the steps in the flowcharts in the accompanying drawings are shown sequentially as indicated by the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless otherwise expressly stated herein, there is no strict order in which these steps are performed, and they may be performed in other orders.

[0382] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A virtual power plant adaptive scheduling method based on health degree credible derating, characterized in that, Includes the following steps: Obtain resource information and external dispatch instructions corresponding to multiple adjustable resources in the virtual power plant; Based on the resource information, resource capacity and constraint modeling is performed to obtain a basic feasible region including the first up and down power boundary. The corresponding health score is calculated based on the communication status and historical performance deviation of each adjustable resource. Based on several of the aforementioned health scores, the basic feasible region is subjected to a credible derating to obtain a committable feasible region that includes the second up and down adjustment power boundaries. Here, the credible derating is characterized by derating the basic adjustability to converge to a credible adjustment range that can be stably fulfilled and does not default. An adaptive margin is calculated based on the health score and prediction error level corresponding to all the adjustable resources. The feasible region is then reduced based on the adaptive margin to obtain an executable feasible region that includes the third up and down power boundary. Based on the external scheduling instructions and the feasible execution domain, a rolling scheduling optimization is constructed to obtain and execute a scheduling scheme for each adjustable resource. Specifically, Configure the window length and the time period that needs to be optimized for each scheduling moment; The resource adjustment amount of each adjustable resource in each time period within the rolling window is used as a decision variable; Construct a quadratic programming objective function to minimize the overall cost. The objective function includes a quadratic term for the corresponding aggregate tracking error and a linear term for the resource usage cost. Within the feasible region, the optimal decision solution that minimizes the objective function is obtained to obtain the optimal adjustment sequence for each time period within the rolling window; Only the optimal adjustment value corresponding to the first time period in the optimal adjustment value sequence is executed and distributed to the corresponding adjustable resource.

2. The virtual power plant adaptive scheduling method based on health-based reliable derating as described in claim 1, characterized in that, The resource information includes the online status, maximum up-adjustment capacity, maximum down-adjustment capacity, ramp rate, energy information, and user service information of each adjustable resource. Based on the resource information, resource capacity and constraint modeling is performed to obtain a basic feasible region including the first up- and down-adjustment power boundaries, including the following steps: The first up-down power boundary is generated based on the maximum up-adjustment capability and the maximum down-adjustment capability combined with the online status. Based on the ramp rate, a ramp rate constraint is generated. When the ramp rate is greater than 0, the absolute difference of the resource adjustment amount between adjacent time points is less than the ramp rate. Based on the energy information, an energy constraint is generated such that when the resource adjustment amount is increased, the corresponding real-time energy decreases and the real-time energy constraint is within a preset energy capacity boundary. Business constraints are generated based on the user business information to ensure that the function value can be calculated and is not greater than 0 after substituting the given resource adjustment amount, the real-time energy and the external business volume into the explicit constraint expression. The resource adjustment amount that satisfies the first power adjustment boundary and all constraints is integrated into the basic feasible domain.

3. The virtual power plant adaptive scheduling method based on health-based reliable derating as described in claim 1, characterized in that, The health score is calculated based on the communication status and historical performance deviation of each adjustable resource, including the following steps: The system acquires online rate, command receipt success rate, packet loss rate, and round-trip latency, performs scaling and normalization, and calculates the communication score at each time point by combining the communication weight set. The relative error is calculated by dividing the deviation amplitude of the target adjustment amount and the actual adjustment amount of each adjustable resource at the same time by the reference power base. The response time of each adjustable resource to the command is obtained. The relative error and the response time are normalized by the exponential decay normalization function to obtain the error sub-score and the time sub-score. The alarm index of the adjustable resource at each time is obtained and normalized to generate an alarm score. The performance score is calculated based on the error sub-score, the time sub-score, and the alarm score, combined with the corresponding performance weight set. The health score is obtained by weighting and fusing the communication score and the performance score, and then using an exponentially weighted moving average for smooth updates.

4. The virtual power plant adaptive scheduling method based on health-based reliable derating as described in claim 3, characterized in that, The basic feasible region is reliably derated based on several of the aforementioned health scores to obtain a committable feasible region that includes a second up-and-down power boundary, including the following steps: The calculated health score is used as a confidence coefficient, and the first up and down power boundary is multiplied by the confidence coefficient to calculate the second up and down power boundary.

5. The virtual power plant adaptive scheduling method based on health degree credible derating according to claim 1, characterized in that, The adaptive margin is calculated based on the health score and prediction error level corresponding to all the adjustable resources, including the following steps: A weighted weight is generated based on the maximum adjustment capability of each adjustable resource, and the health score is combined with the weighted weight to calculate the aggregate health score. The aggregate tracking error is calculated by combining the measured power response data from the previous moment with the external scheduling command from the previous moment and the preset aggregate reference power benchmark. Obtain the smoothing parameters and combine them with the aggregated tracking error. Calculate the prediction error variance at the current time using the exponentially weighted moving average method, and take the square root of the prediction error variance to calculate the prediction error level. The adaptive margin is obtained by multiplying the aggregated health level and the prediction error level by the equilibrium coefficient and summing them.

6. The virtual power plant adaptive scheduling method based on health degree credible derating according to claim 5, characterized in that, It also includes the following steps: The current operating scenario is obtained. If the current operating scenario is a short-term fluctuation, the equilibrium coefficient is increased; if the current operating scenario is a long-term reliable one, the equilibrium coefficient is decreased. The larger the equilibrium coefficient, the more the change in the adaptive margin depends on the change in the prediction error level; the smaller the equilibrium coefficient, the more the change in the adaptive margin depends on the change in the aggregate health.

7. The virtual power plant adaptive scheduling method based on health-reliable derating as described in claim 6, characterized in that, The process of reducing the committable feasible region based on the adaptive margin to obtain an executable feasible region including the third up and down power boundaries includes the following steps: The safety margin coefficient is calculated using the adaptive margin, and the second up and down power adjustment boundary is multiplied by the safety margin coefficient to calculate the third up and down power adjustment boundary.

8. The virtual power plant adaptive scheduling method based on health-based reliable derating as described in claim 1, characterized in that, The quadratic term also includes a tracking priority weight, the larger the value, the higher the system's requirement for tracking accuracy. The linear term also includes a usage coefficient, the larger the value, the more inclined the system is to avoid calling the adjustable resource.

9. The virtual power plant adaptive scheduling method based on health degree credible derating according to claim 1, characterized in that, Also includes: When any of the adjustable resources is detected to meet the preset abnormal triggering conditions, the health score is updated using the fast decay parameter, and the calculation of the second up and down power boundary, the calculation of the adaptive margin, and the solution of the rolling scheduling optimization are re-executed.

10. The virtual power plant adaptive scheduling method based on health degree credible derating according to claim 9, characterized in that, Also includes: If the health score is lower than the removal threshold, then the online status of the adjustable resource is set to 0 and it is removed. After removing the virtual power plants, the maximum upward adjustment capacity and the maximum downward adjustment capacity corresponding to them are recalculated. If the scheduling scheme cannot satisfy the external scheduling instruction, then the external commitment is downgraded to generate a new external scheduling instruction to limit it to the range of aggregated available capabilities.

Images