Virtual power plant transaction scheduling optimization system based on digital twinning

Abstract

The invention relates to the technical field of digital twinning, in particular to a virtual power plant transaction scheduling optimization system based on digital twinning, which comprises a twinning state mapping module, an instruction twinning verification module and a capacity locking distribution module. The twin state mapping module collects distributed resource time sequence data, reconstructs a resource operation efficiency index, and constructs a time sequence bearing graph containing a capacity boundary and an efficiency gradient; the instruction twinning verification module implements time sequence overlay verification of the instruction and the atlas, judges a scheduling conflict time period, shifts an overflow quota time period to a surplus capacity time period, and generates a corrected scheduling task book; and the capacity locking distribution module generates a resource priority sequence based on the response gradient coefficient, locks the node capacity and issues an execution instruction. According to the method, the digital twinborn technology is utilized to perform pre-simulation and time domain optimization of the scheduling plan, the technical problem that external requirements are not matched with physical performance capabilities of the resources is solved, and closed-loop control of resource transaction scheduling is realized.

Description

Technical Field

[0001] This invention relates to the field of digital twin technology, specifically to a virtual power plant trading and scheduling optimization system based on digital twins. Background Technology

[0002] In the field of resource planning and scheduling management, virtual power plants, as digital management units aggregating distributed resources, are essentially complex processes involving the processing of massive amounts of industrial big data. This data encompasses transaction declaration information for heterogeneous resources, millisecond-level real-time telemetry values, equipment health status parameters, and meteorological environmental sensing sequences, exhibiting typical characteristics of multi-source heterogeneity, high-dimensional time-varying nature, and high data throughput characteristic of industrial big data. With the deepening of digital transformation, digital twin technology is gradually being introduced into the operation and management systems of this field. Existing technologies typically utilize big data processing frameworks to clean and store this industrial big data, constructing digital archives of physical assets. This is primarily used to address the "information visualization" problem in resource management, such as generating resource load statistical reports based on historical big data, or monitoring the real-time status of underlying equipment through cloud-based large screens to support basic management oversight and asset inventory.

[0003] However, in the dynamic transaction scheduling and task allocation process for heterogeneous resources, existing technologies suffer from a significant lag in utilizing industrial big data. Existing management systems often rely on static resource quota models or coarse-grained empirical data when formulating scheduling plans, failing to fully leverage the fluctuation characteristics and real-time status patterns inherent in industrial big data. This results in a lack of a mechanism for management systems to dynamically simulate and extrapolate scheduling schemes based on full-scale big data using digital twin models before task issuance. Specifically, the scheduling management layer cannot utilize digital twin technology to extract the temporal value of industrial big data and pre-verify the feasibility of scheduling plans in virtual space, leading to management schemes that often fail to adapt to the actual fulfillment capabilities of resources within specific future time windows.

[0004] This management model, characterized by a disconnect between static planning and dynamic big data, easily leads to the following problems: First, unreasonable task allocation results in the power requirements of target instructions exceeding the dynamic capacity boundaries of resources revealed by big data analysis, leading to task execution failures or low resource fulfillment rates. Second, the lack of refined assessment of resource availability in a timely manner prevents the identification and correction of supply-demand mismatches in the time domain during the scheduling scheme formulation stage, resulting in inefficient resource allocation. In short, existing technologies lack the ability to utilize digital twin models to conduct "pre-simulation and time-series overlay verification" of transaction scheduling instructions based on industrial big data, thus failing to guarantee the accuracy and executability of resource scheduling plans.

[0005] To address this issue, this application proposes a virtual power plant trading and scheduling optimization system based on digital twins. The aim is to solve the problem that the existing scheduling system lacks a simulation and verification process for control commands, which leads to a mismatch between the command sequence and the real-time carrying capacity of physical resources, resulting in large control errors and poor system response stability. Summary of the Invention

[0006] The purpose of this invention is to provide a virtual power plant transaction scheduling optimization system based on digital twins. By constructing a time-series load map and overlaying and verifying it with the target instructions, the system optimizes the time-shift positioning of overflow quotas, thereby achieving time-series scheduling optimization and precise capacity locking allocation for external transaction demands while ensuring the physical constraints of resources.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A virtual power plant trading and scheduling optimization system based on digital twins includes: Twin State Mapping Module: Collects time-series business data of distributed resources in virtual power plants, drives the resource characteristic analysis logic in the digital twin engine to extract data, reconstructs resource operation efficiency indicators and response gradient coefficients, and constructs a time-series load map containing capacity boundaries and efficiency gradients; Command Twin Verification Module: Reads the interaction quota and time window index from the external interaction task package issued by the remote scheduling cloud, and generates the target command sequence; performs time-series overlay verification between the target command sequence and the time-series carrying map in the digital twin space, identifies scheduling conflict periods exceeding the time-series carrying map; extracts the overflow quota within the scheduling conflict period, and locates the surplus capacity period in the time-series carrying map; shifts the overflow quota to the surplus capacity period, applies boundary value constraints to the remaining quota, and generates a corrected scheduling task book; Capacity locking and allocation module: Calculates the performance ranking of virtual nodes based on the reconstructed response gradient coefficient, generates a resource priority sequence, performs descending quota matching on virtual nodes, and calculates task sharing weight; based on the task sharing weight, it decomposes the modified scheduling task into execution scheduling instructions containing physical address information and sends them to the resource execution terminal.

[0008] Preferably, the process of the twin state mapping module collecting and extracting time-series business data includes: collecting historical transaction declaration sequences and real-time power telemetry values ​​uploaded by the virtual power plant's distributed resources as time-series business data; transmitting the time-series business data to the virtual mapping layer of the digital twin engine through the data acquisition interface; mapping the time-series business data onto the time axis using the time-series alignment logic built into the digital twin engine; and constructing a resource operation basic dataset by performing cleaning and alignment processing on the mapped data through the digital twin engine.

[0009] Preferably, the process of reconstructing the resource operation performance index and response gradient coefficient by the twin state mapping module includes: reading the resource operation basic dataset output by the digital twin engine; calculating the synchronization absolute difference of the time-series business data and performing accumulation and averaging processing to obtain the resource operation performance index; extracting the power regulation rate value and operation stability value from the resource operation basic dataset; and using the weighted calculation module built into the digital twin engine to perform normalized weighted processing on the resource operation performance index, power regulation rate value and operation stability value to generate the response gradient coefficient.

[0010] Preferably, the process of constructing the time-series capacity map by the twin state mapping module includes: extracting physical constraint parameters from the resource operation basic dataset, extrapolating the upper and lower limits of output within a preset time period using a digital twin engine, constructing capacity boundaries using the upper and lower limits of output, and generating a closed capacity envelope domain on the time axis; mapping the response gradient coefficient as a weight value to the coordinate space of the capacity envelope domain, and using the distribution of weight values ​​in the coordinate space as the efficiency gradient to generate a time-series capacity map containing the capacity boundary and the efficiency gradient.

[0011] Preferably, the process of the instruction twin verification module generating the target instruction sequence includes: reading the external interactive task package transmitted from the remote scheduling cloud, extracting the interactive quota containing the total task requirements, and reading the target regulation power value from the interactive quota; extracting the time window index containing the task execution period, and reading the task start and end timestamps from the time window index; obtaining the simulation time step parameter in the digital twin engine; using the simulation time step parameter to perform time-domain segmentation on the time range limited by the task start and end timestamps to generate discrete time nodes; mapping the target regulation power value to the discrete time nodes and generating instruction power values ​​corresponding to each discrete time node; and constructing the target instruction sequence based on the discrete time nodes and instruction power values.

[0012] Preferably, the process of the instruction twin verification module in identifying scheduling conflict periods includes: mapping discrete time nodes in the target instruction sequence to a time-series carrying capacity map and aligning them on the time axis; comparing the instruction power values ​​at each discrete time node with the upper limit threshold of the capacity envelope domain in the time-series carrying capacity map; marking discrete time nodes with instruction power values ​​greater than the upper limit threshold of the capacity envelope domain as scheduling conflict periods; calculating the difference between the instruction power values ​​and the upper limit threshold of the capacity envelope domain within the scheduling conflict period to obtain the overflow quota value; and marking discrete time nodes with instruction power values ​​less than the upper limit threshold of the capacity envelope domain as surplus capacity periods.

[0013] Preferably, the process of the instruction twin verification module generating the revised scheduling task book includes: retrieving surplus capacity periods in the time-series load map based on the time axis index, where the time interval between the scheduling conflict period and the conflict period is within a preset threshold; calculating the difference between the upper limit threshold of the capacity envelope domain and the instruction power value of the surplus capacity period to obtain the allowable margin value; accumulating the overflow quota value into the instruction power value of the surplus capacity period, and performing numerical limiting processing on the accumulation result using the upper limit threshold of the capacity envelope domain; resetting the instruction power value of the scheduling conflict period to the upper limit threshold of the capacity envelope domain, and performing numerical truncation processing on the overflow quota value exceeding the allowable margin value; and integrating the instruction power values ​​of each discrete time node after resetting and accumulation processing to generate the revised scheduling task book.

[0014] Preferably, the process of generating a resource priority sequence by the capacity locking allocation module includes: extracting the device identification code of each virtual node in the digital twin space, and reading the response gradient coefficient associated with the device identification code; using the numerical sorting logic built into the digital twin engine, performing a descending sorting process on the device identification code based on the value of the response gradient coefficient; assigning a call order index to each device identification code according to the result of the descending sorting process; and constructing a resource priority sequence based on the device identification code and the call order index.

[0015] Preferably, the process by which the capacity locking allocation module generates and sends execution scheduling instructions to the terminal includes: reading the instruction power value and the call order index in the resource priority sequence from the modified scheduling task book; extracting the current available capacity of each virtual node according to the call order index; performing a step-by-step deduction calculation on the current available capacity of each virtual node using the instruction power value until the instruction power value is zero, thereby determining the actual task load of each virtual node; retrieving the physical network address associated with each virtual node; encapsulating the task load value and the physical network address to generate an execution scheduling instruction containing the execution action; and sending the execution scheduling instruction to the corresponding resource execution terminal through the communication interface.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Based on digital twin technology, this invention uses multi-dimensional data such as synchronous absolute difference and power regulation rate to normalize and weight the data. By mapping the response gradient coefficient to the efficiency gradient, a time-series load map containing capacity boundary and efficiency gradient is obtained. This solves the problem of the existing technology's single characterization of resource capacity and insufficient dynamism, and realizes the faithful mapping and reliable modeling of the virtual power plant's resource operation status.

[0017] 2. This invention utilizes a time-series overlay verification mechanism within a digital twin space. By calculating overflow quotas and surplus capacity periods, and retrieving time intervals within a preset threshold and performing time-shift positioning optimization, a revised scheduling task book is obtained. This solves the problem that capacity conflicts in traditional scheduling can only be simply truncated, and realizes automatic and intelligent optimization and shifting of external transaction demands on the time axis to optimize the accommodation of scheduling demands.

[0018] 3. This invention adopts performance ranking and priority construction based on response gradient coefficient. By performing step-by-step deduction calculations on the current available capacity of resources through the order of calls, the actual task sharing value of each virtual node is obtained. This solves the problem of inefficient resource utilization caused by traditional allocation mechanisms, realizes the calculation of task sharing weight and accurate locking of capacity, and ensures the execution rate of scheduling instructions. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the structure of a virtual power plant trading and scheduling optimization system based on digital twins according to the present invention; Figure 2 This is a schematic diagram of capacity boundaries and instruction stacking according to an embodiment of the present invention; Figure 3 This is a flowchart illustrating the instruction twin verification and scheduling correction logic of an embodiment of the present invention. Detailed Implementation

[0020] 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figures 1 to 3 This invention provides a virtual power plant trading and scheduling optimization system based on digital twins, the technical solution of which is as follows: Example

[0022] Reference Figure 1 This is a schematic diagram of the structure of a virtual power plant trading and scheduling optimization system based on digital twins according to the present invention. This embodiment provides an application scenario for the virtual power plant trading and scheduling optimization system based on digital twins. The specific application scenario is a regional virtual power plant that integrates distributed photovoltaic power stations, energy storage power stations, and controllable loads. The specific steps include: Twin State Mapping Module: Collects time-series business data of distributed resources in virtual power plants, drives the resource characteristic analysis logic in the digital twin engine to extract data, reconstructs resource operation efficiency indicators and response gradient coefficients, and constructs a time-series load map containing capacity boundaries and efficiency gradients; Command Twin Verification Module: Reads the interaction quota and time window index from the external interaction task package issued by the remote scheduling cloud, and generates the target command sequence; performs time-series overlay verification between the target command sequence and the time-series carrying map in the digital twin space, identifies scheduling conflict periods exceeding the time-series carrying map; extracts the overflow quota within the scheduling conflict period, and locates the surplus capacity period in the time-series carrying map; shifts the overflow quota to the surplus capacity period, applies boundary value constraints to the remaining quota, and generates a corrected scheduling task book; Capacity locking and allocation module: Calculates the performance ranking of virtual nodes based on the reconstructed response gradient coefficient, generates a resource priority sequence, performs descending quota matching on virtual nodes, and calculates task sharing weight; based on the task sharing weight, it decomposes the modified scheduling task into execution scheduling instructions containing physical address information and sends them to the resource execution terminal.

[0023] Furthermore, the process of the twin state mapping module collecting and extracting time-series business data includes: collecting historical transaction declaration sequences and real-time power telemetry values ​​uploaded by the virtual power plant's distributed resources as time-series business data; transmitting the time-series business data to the virtual mapping layer of the digital twin engine through the data acquisition interface; mapping the time-series business data onto the time axis using the time alignment logic built into the digital twin engine; and constructing a resource operation basic dataset by performing cleaning and alignment processing on the mapped data through the digital twin engine.

[0024] Specifically, the digital twin state mapping module utilizes an industrial IoT technology architecture to capture massive amounts of heterogeneous industrial big data from multiple sources in real time through edge computing gateways or smart terminals deployed at various physical resource endpoints. This data constructs a time-series feature space: historical transaction declaration sequences include regulatory capacity quotation data and power output plan data reported by each resource within a preset period; real-time power telemetry values ​​are generated from high-frequency sampling of the current active power output of photovoltaic inverters, state-of-charge values ​​uploaded by the battery management system (BMS), and real-time power consumption values ​​of controllable loads. The data acquisition interface uses standardized industrial transmission protocols to transmit these heterogeneous data streams to the virtual mapping layer of the digital twin engine with high throughput.

[0025] To address the typical multi-granularity characteristics of industrial big data (e.g., energy storage devices are sampled at the second level, while photovoltaic devices are sampled at the minute level), the digital twin engine activates its built-in time-series alignment logic, setting a unified baseline time step to achieve data synchronization. For data with a sampling frequency higher than the baseline step, the system performs downsampling processing based on the mean within the time step to reduce data dimensionality. For data with a sampling frequency lower than the baseline step, the system uses an interpolation algorithm to fill in time gaps. This interpolation algorithm is based on a linear relationship, calculating the data value of the target time point by using two adjacent known valid time points on either side of the target time point and their corresponding values. Specifically, assuming the time t to be filled is between known time points t1 and t2, the data value at time t is calculated based on the proportional relationship of the time intervals. Subsequently, the system performs ETL (Extract, Transform, Load) cleaning and data quality governance operations on the mapped data, removing null values ​​caused by communication packet loss (replacing them with the previous valid value), and identifying and removing noisy data that significantly deviates from physical characteristics. Noise identification is based on the statistical principle of three standard deviations (3σ): for a dataset within a time step, if the value of any data point deviates from the mean of the dataset by more than three standard deviations, it is judged as abnormal noise and replaced with the median of the adjacent time step, ultimately generating a standardized resource operation basic dataset.

[0026] This invention solves the problem of asynchronous data in the time dimension of multi-source heterogeneous devices by collecting time-series business data and utilizing the time alignment logic and cleaning processing built into the digital twin engine. It eliminates the data barriers caused by different sampling frequencies and communication protocols, and provides a standardized and reliable data foundation for subsequent state extrapolation and simulation in the digital twin space.

[0027] Furthermore, the process by which the twin state mapping module reconstructs the resource operation performance index and response gradient coefficient includes: reading the resource operation basic dataset output by the digital twin engine; calculating the synchronization absolute difference of the time-series business data and performing accumulation and averaging processing to obtain the resource operation performance index; extracting the power regulation rate value and operation stability value from the resource operation basic dataset; and using the weighted calculation module built into the digital twin engine to perform normalized weighted processing on the resource operation performance index, power regulation rate value, and operation stability value to generate the response gradient coefficient.

[0028] Specifically, in this embodiment, the twin state mapping module retrieves the cleaned and aligned resource runtime base dataset from the storage unit of the digital twin engine. The system selects the planned power value P from the historical transaction declaration sequence. plan (t) i ) and real-time power telemetry value P real (t) iAs the computational object, at each aligned time node t i The absolute value of the difference between the planned power and the real-time power is calculated, which is the synchronization absolute difference. This value physically represents the performance deviation of the distributed resource at a specific moment. The system performs an accumulation operation on all synchronization absolute differences within a set period and divides the accumulation result by the total number of time nodes N, thereby quantifying the resource operation efficiency index E. The calculation process of the resource operation efficiency index E is as follows: sum the absolute values ​​of the difference between the planned power and the real-time power at each time node within the set period, and then divide the sum by the total number of time nodes. The smaller the value of this index, the higher the historical performance accuracy of the resource.

[0029] The specific process of reconstructing resource operation performance indicators includes: adjusting the set period length used for accumulation and averaging processing based on the operational status change trend of time-series business data in the resource operation basic dataset; when the resource operation status change trend exceeds a preset fluctuation threshold, shortening the set period length to generate resource operation performance indicators. When calculating resource operation performance indicator E, the system first analyzes the synchronous absolute difference sequence in the historical transaction declaration sequence to calculate the operational status change trend of the sequence. If the absolute value of this change rate exceeds a preset fluctuation threshold (e.g., 10%) for three consecutive days, it is determined that the resource operation status has changed significantly, and the system dynamically shortens the set period length used to calculate indicator E from the original 24 hours to 12 hours. This dynamic adjustment mechanism ensures that indicator E can quickly reflect the latest operational performance of the resource. This scheme achieves adaptive adjustment of the performance indicator calculation window, improving the accuracy of the response gradient coefficient in evaluating the current performance of the resource.

[0030] Simultaneously, the ramp rate of the energy storage power station (as the power regulation rate value R, in kW / min) is extracted, and this rate is defined as the ramp rate between two adjacent aligned time steps. Power change within ,Right now The system also includes the voltage fluctuation variance of the photovoltaic inverter (as the operational stability value S), which is calculated based on the standard statistical variance of voltage samples over the past hour. The digital twin engine constructs a three-dimensional feature vector containing the resource operational efficiency index E, the power regulation rate value R, and the operational stability value S. It uses a range transformation method to map these three dimensions-different values ​​to a dimensionless interval of 0 to 1. The maximum and minimum values ​​used in the range transformation method are based on the historical statistical extreme values ​​of all similar resources over the past year. Subsequently, the system uses preset strategy weights (efficiency index weights)... Adjustment rate weight Stability weights For the normalized values The weighted summation calculation yields the scalar value of the response gradient coefficient. The calculation formula is: .

[0031] This invention calculates and normalizes the historical performance, adjustment rate, and operational stability indicators of resources, transforming various physical characteristics and historical behavior data of heterogeneous resources into a unified quantitative evaluation standard. This eliminates the measurement differences in adjustment capacity and performance quality among different types of resources, providing a quantitative and reliable coefficient basis for constructing a carrying capacity map with real physical attributes in the digital twin space.

[0032] Furthermore, the process of constructing the time-series capacity map by the twin state mapping module includes: extracting physical constraint parameters from the resource operation basic dataset, extrapolating the upper and lower limits of output within a preset time period using a digital twin engine, constructing capacity boundaries using the upper and lower limits of output, and generating a closed capacity envelope domain on the time axis; mapping the response gradient coefficient as a weight value to the coordinate space of the capacity envelope domain, and using the weight value distribution in the coordinate space as the efficiency gradient to generate a time-series capacity map containing the capacity boundary and the efficiency gradient.

[0033] Specifically, in this embodiment, the system first extracts the physical constraint parameters from the resource operation basic dataset, including the upper and lower limits of the state of charge (SOC) of the energy storage power station, the extreme values ​​of charging and discharging power, and the rated capacity of the distributed photovoltaic inverter. The digital twin engine, based on meteorological forecast data and the equipment physical model, extrapolates the upper and lower limits of output for each resource node at each time step within a preset future time period. The extrapolation model for energy storage equipment is based on the energy balance equation of the SOC and the charging and discharging efficiency matrix; the extrapolation model for photovoltaic equipment is based on the nonlinear empirical curve of irradiance-temperature-power. For energy storage equipment, the future extrapolation of the SOC is achieved by performing time-cumulative calculations on the predicted charging and discharging power under the current SOC and substituting this into the equipment's charging and discharging efficiency and rated capacity parameters. This extrapolation logic ensures that the equipment will not exceed its physical energy storage limit during operation. For example, for energy storage equipment, when the SOC approaches its upper limit, the upper limit of output in the charging direction is compressed to zero by the extrapolation logic; for photovoltaic equipment, its upper limit of output fluctuates dynamically with the irradiance prediction curve. The system connects the aforementioned time-varying upper and lower thresholds to construct a capacity boundary in a time-power two-dimensional coordinate system, thereby defining a closed capacity envelope. This capacity envelope physically represents the entire feasible adjustment range of the virtual power plant in the future. Logically, this capacity envelope is decomposed into time-shiftable capacity (elastic adjustment quantity) provided by flexible resources such as energy storage and controllable loads, and non-time-shiftable capacity (predictive baseline quantity) provided by uncontrollable resources such as photovoltaics and wind power. Subsequently, the system reads the response gradient coefficient G generated in the preceding steps and uses it as a weight value representing the quality of resource adjustment. The system uniformly assigns this weight value to all time-power coordinate points within the capacity envelope, ensuring that all feasible adjustment quantities within the envelope carry the same adjustment efficiency value attribute. Thus, the distribution of weight values ​​in the coordinate space constitutes the efficiency gradient. The data structure of this efficiency gradient is a three-dimensional time-power-efficiency tensor. The final generated time-series carrying capacity map is a multi-dimensional data model integrating the physical feasible domain (capacity boundary) and the adjustment efficiency distribution (efficiency gradient). The efficiency gradient is mainly reflected in the dynamic changes on the time axis: since the response gradient coefficient G is a dynamic value that changes with time, there are significant differences in the efficiency values ​​within the capacity envelope at different time slices, thus forming an efficiency gradient that can be optimized in the time dimension.

[0034] This invention constructs a digital map that combines physical hard constraints and performance soft indicators by deriving dynamic capacity boundaries and embedding response gradient coefficients as efficiency gradients. It not only intuitively defines the regulation capacity boundary of virtual power plants, but also realizes the quantitative stratification of regulation resource quality at different time periods, providing a quantitative decision-making basis with multi-dimensional characteristics for the matching of subsequent dispatch instructions in time and space.

[0035] Furthermore, the process by which the instruction twin verification module generates the target instruction sequence includes: reading the external interactive task package transmitted from the remote scheduling cloud, extracting the interactive quota containing the total task requirements, and reading the target regulation power value from the interactive quota; extracting the time window index containing the task execution period, and reading the task start and end timestamps from the time window index; obtaining the simulation time step parameter in the digital twin engine; using the simulation time step parameter to perform time-domain segmentation of the time range limited by the task start and end timestamps to generate discrete time nodes; mapping the target regulation power value to the discrete time nodes and generating instruction power values ​​corresponding to each discrete time node; and constructing the target instruction sequence based on the discrete time nodes and instruction power values.

[0036] Specifically, the system first receives an external interactive task packet from the remote dispatch cloud via a dedicated network or encrypted internet communication interface. The system parses this task packet, identifies the interactive quota field, and reads the target adjustment power value for this dispatch request. Simultaneously, the module parses the time window index field to determine the task's effective time range, i.e., the task's start and end timestamps. Next, the module calls the configuration parameters of the digital twin engine to obtain the preset simulation time step parameter. The system then uses this simulation time step... Time range for task start and end timestamps Perform time-domain segmentation to generate discrete time node sequence t i The segmentation logic follows these rules: start timestamp It must be included in the sequence; subsequent time node t i+1 equal to t i +Δt, until t i Equal to or greater than the end timestamp Until then, and It must also be included in the sequence. Subsequently, the system assigns the target adjustment power value to each discrete time node, generating the instruction power value at the corresponding time point. Finally, the system combines these discrete time nodes and the corresponding instruction power values ​​in a time sequence to construct a target instruction sequence that can be read and executed point by point in the digital twin space.

[0037] This invention transforms macroscopic scheduling task packages into discrete instruction sequences adapted to the computational granularity of digital twin engines by dividing and discretizing them in the time domain based on the simulation time step. This achieves precise alignment between external scheduling requirements and the internal simulation environment in the time domain, providing a precisely aligned data structure for subsequent time-series overlay verification at the micro-time scale.

[0038] Furthermore, the process by which the instruction twin verification module identifies scheduling conflict periods includes: mapping discrete time nodes in the target instruction sequence to a time-series carrying capacity map and aligning them on the time axis; comparing the instruction power values ​​at each discrete time node with the upper limit threshold of the capacity envelope domain in the time-series carrying capacity map; marking discrete time nodes with instruction power values ​​greater than the upper limit threshold of the capacity envelope domain as scheduling conflict periods; calculating the difference between the instruction power value and the upper limit threshold of the capacity envelope domain within the scheduling conflict period to obtain the overflow quota value; and marking discrete time nodes with instruction power values ​​less than the upper limit threshold of the capacity envelope domain as surplus capacity periods. (Reference) Figure 2 This is a schematic diagram of capacity boundaries and instruction stacking in an embodiment of the present invention.

[0039] Specifically, the system precisely overlays the generated discrete time node sequence containing specific power requirements onto the coordinate system of the time-series capacity map in the time dimension, ensuring that the two sets of data are strictly aligned at the same timestamp. The system reads the commanded power value of each discrete time node point by point and extracts the upper limit threshold of the capacity envelope domain corresponding to that time point in the time-series capacity map. If the system determines that the commanded power value is greater than the upper limit threshold, it marks that time as a scheduling conflict period. Subsequently, the system performs difference calculation to obtain the gap for that period, defines it as the overflow quota value, and temporarily stores it. Conversely, if the commanded power value is less than the upper limit threshold, the system determines that the commanded power value is less than the upper limit threshold and marks that time node as a surplus capacity period, indicating that that period has the ability to absorb additional tasks. The verification and overflow quota calculation are evaluated based on the controllable regulation capability of the virtual power plant after deducting the non-time-shiftable capacity, ensuring that the overflow quota is quantified only for the gap in the flexible resource (time-shiftable capacity) part.

[0040] This invention compares the upper limit threshold of the target instruction with the capacity of the timing load map, and utilizes the advanced simulation capability of the digital twin space to accurately identify the timing mismatch points between external demand and internal physical capacity before the task is issued, and quantifies the power gap that is not executable. This provides quantitative spatiotemporal verification results and correction basis for subsequent scheduling task time shift correction and peak shaving and valley filling optimization.

[0041] Furthermore, the process of the instruction twin verification module generating the revised scheduling task book includes: retrieving surplus capacity periods in the time-series load map based on the time axis index, where the time interval between scheduling conflict periods is within a preset threshold; calculating the difference between the upper limit threshold of the capacity envelope domain and the instruction power value of the surplus capacity period to obtain the allowable margin value; accumulating the overflow quota value into the instruction power value of the surplus capacity period, and performing numerical limiting processing on the accumulation result using the upper limit threshold of the capacity envelope domain; resetting the instruction power value of the scheduling conflict period to the upper limit threshold of the capacity envelope domain, and performing numerical truncation processing on the overflow quota value exceeding the allowable margin value; integrating the instruction power values ​​of each discrete time node after reset and accumulation processing to generate the revised scheduling task book. (Reference) Figure 3 This is a flowchart of the instruction twin verification and scheduling correction logic in an embodiment of the present invention.

[0042] Specifically, the system uses the time node marked as a scheduling conflict period (e.g., time T1, overflow quota is...) Centered on a timeline index, it performs bidirectional searches for surplus capacity periods within a preset range. Assuming the search finds... During the period of surplus capacity, the command power value at this time is The upper limit threshold of the capacity envelope domain shown by the time-series carrying capacity map is... The system calculates the difference between the two. This value is defined as the allowable margin value. Subsequently, the system performs a time-shift operation for directional matching: if To increase power demand, adjustments will only be made when there is a margin for capacity increase. Location is determined by time period; if To reduce power demand, only when there is a capacity reduction margin will it be implemented. Positioning is done within a specific time period. The overflow quota value at time T1 is allocated to... The corrected value in the command power value at time is At this point, the system checks again whether the correction value exceeds the upper limit threshold at that moment. If it does, it performs numerical limiting processing, forcibly clamping the value to the upper limit threshold. For conflict source T1, the system forcibly resets the instruction power value at that moment to the physical limit value (i.e., the upper limit threshold of the capacity envelope). If the total capacity of all surplus periods within the search range is still less than the overflow quota, the system performs numerical truncation on the remaining untransferable overflow portion, i.e., discards it directly. Finally, the system re-integrates all discrete time node data after reset (peak shaving) and accumulation (valley filling) processing to generate a corrected scheduling task book that both conforms to physical constraints and retains as much of the total task volume as possible.

[0043] The specific process of shifting the overflow quota to the surplus capacity period includes: based on the efficiency gradient in the time-series carrying capacity map, selecting target periods within the surplus capacity period where the efficiency gradient is higher than a preset threshold and the capacity boundary volatility is lower than a preset threshold; calculating the compatibility between the overflow quota and the remaining carrying capacity of the target period to generate a spatiotemporal matching score sequence; and decomposing and shifting the overflow quota to the target period according to the score sequence to generate a revised scheduling task book.

[0044] Specifically, when it is determined that there is a problem in time period T1 After the quota overflows, the system first scans the preset range ( Efficiency gradient within the surplus capacity period. The system uses capacity boundaries. The rate of change on the time axis (i.e.) ) to perform calculations, where This represents the change in the capacity boundary within the current time step. For a preset time step, this rate of change characterizes the volatility of the capacity boundary. The system filters out... China satisfies and Target period ,(in The preset efficiency gradient threshold, (A preset boundary volatility threshold) is used to ensure that quotas are allocated to periods with high adjustment efficiency and stable operation (small boundary volatility). Subsequently, the system will... Treat it as a task package, calculate its relationship with all The adaptability of remaining capacity for a given time period can be assessed using algorithms such as linear programming (LP) or mixed-integer programming (MIP), with the objective function being minimizing the weighted cost of the time-shift quota. This weighted cost function is determined by the time-shift distance cost (…). The constraints are weighted by the time interval between the target time period and time period T1, and the efficiency gradient loss cost (the difference between the G value of the target time period and the G value of time period T1). In addition to total balance constraints and capacity limits, the constraints also include power ramp-up rate constraints for each resource to ensure that the power change rate after allocation does not exceed the physical adjustment limit of the equipment in adjacent time steps. The constraint is that the sum of all time-shifted decomposition quotas equals... And each allocation quota shall not exceed The system calculates the available capacity for each time period and generates an optimized spatiotemporal matching scheme. Finally, the system uses this scheme to... Decompose into P a P b Sub-quotas are allocated and shifted to the optimal target time period T. a T bThis allows for the efficient generation of revised scheduling task books. The revised scheduling task book represents the virtual power plant's final, time-optimal performance commitment to the power grid based on its own physical constraints and optimization strategies. After generating the revised scheduling task book, the system uses a digital twin engine to perform dynamic re-simulation of time-coupled resources such as energy storage. This simulation aims to verify whether the revised task sequence will lead to excessive State of Charge (SOC) in subsequent periods. If an excessive limit is found, the system will mark the task book as infeasible and return to the time-shift optimization logic, iteratively solving by adjusting constraints or weights until the time-coupling constraints are met. This scheme, by integrating efficiency gradients and capacity boundary volatility, achieves intelligent reconfiguration of overflow quotas to stable periods, avoiding the secondary performance risk caused by simple shifting, and improving the execution reliability and final performance accuracy of scheduling tasks.

[0045] This invention implements flexible reconfiguration of scheduling tasks based on multidimensional information of time-series load maps. Under the premise of ensuring that scheduling instructions never exceed physical safety boundaries, it maximizes the response rate of virtual power plants to grid scheduling tasks through load shifting in the time dimension, and avoids the loss of traded electricity caused by hard load shedding.

[0046] Furthermore, the process of generating a resource priority sequence by the capacity locking allocation module includes: extracting the device identification code of each virtual node in the digital twin space, and reading the response gradient coefficient associated with the device identification code; using the numerical sorting logic built into the digital twin engine, performing a descending sorting process on the device identification code based on the value of the response gradient coefficient; assigning a call order index to each device identification code according to the result of the descending sorting process; and constructing a resource priority sequence based on the device identification code and the call order index.

[0047] Specifically, the capacity locking and allocation module first traverses the resource topology tree in the digital twin space, extracting all virtual nodes participating in this scheduling and their unique identifiers, i.e., device identifiers. Then, the system retrieves the response gradient coefficients for each node calculated in the previous steps from the database.

[0048] The specific process of sorting device identifiers in descending order based on the value of the response gradient coefficient includes: reading the historical call frequency of virtual nodes and calculating a dynamic penalty term; applying the dynamic penalty term to the response gradient coefficient to generate a modified response gradient coefficient; and sorting the device identifiers in descending order based on the modified response gradient coefficient to generate a resource priority sequence.

[0049] Specifically, before sorting in descending order, the system first reads the historical call frequency of each virtual node over the past week, based on the actual number of tasks it has been assigned. The system calculates a dynamic penalty term based on this frequency. ,in Where K is the penalty weight coefficient. Subsequently, the system calculates... To generate corrected response gradient coefficients ,in This is the original response gradient coefficient. This correction coefficient effectively reduces the resources that have been frequently invoked recently. (Higher) priority. Ultimately, the digital twin engine is based on... The device identifiers are sorted in descending order to generate a balanced and optimized resource priority sequence. This scheme achieves adaptive fine-tuning of priorities based on historical behavior data, balancing the goals of efficiency optimization and fair resource utilization, effectively extending resource lifespan and improving the overall resource resilience of the virtual power plant.

[0050] The digital twin engine initiates a numerical sorting logic, arranging the aforementioned coefficients in descending order of their values. Based on this arrangement, the system assigns a call sequence index to each device identifier. Finally, the system packages the device identifier and its corresponding call sequence index into an ordered list data structure, i.e., a resource priority sequence, establishing the scheduling principle of "prioritizing high-quality resources."

[0051] This invention constructs a dynamic optimization mechanism based on comprehensive resource efficiency by sorting resource identifiers in descending order according to response gradient coefficients and building a priority sequence. This ensures that resources with fast response speed and stable operation are prioritized when executing scheduling tasks, thereby guaranteeing the execution accuracy and fulfillment reliability of the entire virtual power plant to external scheduling commands at the physical level.

[0052] Furthermore, the process by which the capacity locking allocation module generates and sends execution scheduling instructions to the terminal includes: reading the instruction power value and the call order index in the resource priority sequence from the modified scheduling task book; extracting the current available capacity of each virtual node according to the call order index; performing a step-by-step deduction calculation on the current available capacity of each virtual node using the instruction power value until the instruction power value is zero, thereby determining the actual task load of each virtual node; retrieving the physical network address associated with each virtual node; encapsulating the task load value and the physical network address to generate an execution scheduling instruction containing the execution action; and sending the execution scheduling instruction to the corresponding resource execution terminal through the communication interface.

[0053] Specifically, the system first reads the commanded power value for a specific time node in the correction scheduling task book. Simultaneously, the system calls the resource priority sequence generated in the previous steps. Based on the call order index "1", the system first accesses the first-ranked virtual node and extracts its current available capacity. Before performing the step-by-step deduction calculation, the system calls the distribution network topology model and real-time operating status data from the digital twin engine to perform power flow and voltage constraint verification on the virtual node's current available capacity. If the capacity limit would cause local lines to exceed limits, the system performs numerical limiting processing on the limit to generate an "allocated capacity limit" that meets the grid's safe operation constraints. The system performs the step-by-step deduction calculation: deducting the node's allocated capacity limit from the total commanded demand to determine the node's actual task share. Subsequently, the system continues to access the virtual node with the call order index "2" and extracts its available capacity. Since the remaining demand is less than the node's available capacity, the system allocates the remaining demand to the node in full, determining its actual task share. At this point, the commanded power value is zero, the calculation terminates, and nodes with subsequent priorities will not undertake this task. After allocation, the system retrieves the physical network addresses corresponding to the aforementioned nodes from the device configuration library, encapsulates the determined task distribution values ​​with the corresponding addresses using a protocol, and generates execution scheduling instructions that conform to industrial control standards and contain specific adjustment actions. Finally, the system accurately sends these instructions to the resource execution terminals located in the physical field through the IoT communication interface.

[0054] This invention, through the above technical solution, transforms the optimization strategy in the digital twin space into executable signals in the physical world. By using a "greedy algorithm"-like step-by-step deduction logic, it ensures that high-quality resources with high response gradient coefficients are prioritized, thus achieving effective closed-loop control from virtual simulation to physical execution.

[0055] Example 2 This embodiment provides a virtual power plant transaction scheduling optimization system based on digital twins, physically deployed on a virtual power plant management cloud platform. The system's underlying hardware includes a high-performance computing server, a time-series database server, and IoT communication gateways distributed at the edge. Each functional module runs as a microservice component within the computing server, interacting with data via an internal bus. The specific workflow is as follows: Phase 1: Construction of the digital twin foundation (executed by the twin state mapping module).

[0056] The system first activates the digital twin state mapping module, which acts as the system's data acquisition front-end. This module acquires distributed resource data (such as photovoltaics, energy storage, and air conditioning) from the physical layer in real time through an IoT communication gateway. It periodically captures all business data and, using the pre-set device physical models and state deduction algorithms in the digital twin engine, transforms the discrete, heterogeneous raw data into a standardized "time-series capacity map." In this map, capacity boundaries clearly define the physical limits of the system at each moment, while efficiency gradients quantify the system's economic and performance indicators at each moment.

[0057] Phase 2: Scheduling instruction simulation and correction (executed by the instruction twin verification module).

[0058] When the remote dispatch cloud issues external interactive task packages (such as power grid peak-shaving instructions), the instruction twin verification module does not directly forward the instructions, but first performs simulation verification in the digital twin space. The module projects the external instruction sequence onto the aforementioned "time-series load map". The system automatically identifies whether the instruction curve exceeds the capacity boundary. If the instruction is within the capacity limit, the verification will pass.

[0059] If the number of instructions exceeds the capacity limit during a certain period (i.e., a scheduling conflict period), the system searches for nearby surplus capacity periods. The system uses the time-domain extrapolation capability of digital twins to "transfer" the overflow quota from the conflict period to the surplus period, generating a corrected scheduling task book that has been optimized by time-series translation, while ensuring that the physical limits are not exceeded.

[0060] Phase 3: Precise resource locking and distribution (executed by the capacity locking and allocation module).

[0061] After the revised scheduling task list is generated, the capacity locking allocation module initiates the task distribution logic. This module prioritizes all available resources based on "efficiency gradient," generating a resource priority sequence. The system employs an efficiency gradient-based allocation strategy, prioritizing the allocation of task metrics to high-quality resources with high response gradient coefficients. After calculating the specific workload distribution for each node, the system encapsulates this into low-level control commands and sends them to physical terminals for execution via the communication interface.

[0062] This embodiment, through the closed-loop process of the above three stages, completes the mapping of physical resources, the simulation and verification of external commands, and the optimized allocation of execution actions in the digital twin space. This system effectively solves the problem of resource response failure or equipment damage caused by blindly issuing commands in traditional scheduling methods, ensuring that when virtual power plants participate in grid transactions, they can maximize the absorption of scheduling quotas while guaranteeing the operational safety and efficiency of underlying assets.

[0063] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A virtual power plant trading and scheduling optimization system based on digital twins, characterized in that, include: Twin State Mapping Module: Collects time-series business data of distributed resources in virtual power plants, drives the resource characteristic analysis logic in the digital twin engine to extract data, reconstructs resource operation efficiency indicators and response gradient coefficients, and constructs a time-series load map containing capacity boundaries and efficiency gradients; Instruction twin verification module: Reads the interaction quota and time window index from the external interaction task package issued by the remote scheduling cloud, and generates the target instruction sequence; In the digital twin space, perform time-series overlay verification of the target instruction sequence and the time-series bearer map to identify scheduling conflict periods that exceed the time-series bearer map; Extract the overflow quota during the scheduling conflict period and locate the surplus capacity period in the time-series load map; The overflow quota is shifted to the surplus capacity period, the remaining quota is subject to boundary numerical constraints, and a revised scheduling task book is generated. Capacity locking and allocation module: Calculates the performance ranking of virtual nodes based on the reconstructed response gradient coefficient, generates a resource priority sequence, performs descending quota matching on virtual nodes, and calculates task sharing weight. Based on the task sharing weight, the revised scheduling task book is decomposed into execution scheduling instructions containing physical address information and sent to the resource execution terminal.

2. The virtual power plant trading and scheduling optimization system based on digital twins according to claim 1, characterized in that, The process of the twin state mapping module collecting and extracting time-series business data includes: collecting historical transaction declaration sequences and real-time power telemetry values ​​uploaded by the virtual power plant's distributed resources as time-series business data; transmitting the time-series business data to the virtual mapping layer of the digital twin engine through the data acquisition interface; mapping the time-series business data onto the time axis using the time alignment logic built into the digital twin engine; and constructing a resource operation basic dataset by performing cleaning and alignment processing on the mapped data through the digital twin engine.

3. The virtual power plant trading and scheduling optimization system based on digital twins according to claim 1, characterized in that, The process of reconstructing resource operation performance indicators and response gradient coefficients by the twin state mapping module includes: reading the resource operation basic dataset output by the digital twin engine; calculating the synchronization absolute difference of time-series business data and performing accumulation and averaging to obtain resource operation performance indicators; extracting power regulation rate values ​​and operation stability values ​​from the resource operation basic dataset; and using the weighted calculation module built into the digital twin engine to perform normalized weighted processing on the resource operation performance indicators, power regulation rate values, and operation stability values ​​to generate response gradient coefficients.

4. The virtual power plant trading and scheduling optimization system based on digital twins according to claim 1, characterized in that, The process of constructing the time-series capacity map by the twin state mapping module includes: extracting physical constraint parameters from the resource operation basic dataset; using a digital twin engine to extrapolate the upper and lower limits of output within a preset time period; constructing capacity boundaries using the upper and lower limits of output; and generating a closed capacity envelope domain on the time axis; mapping the response gradient coefficients as weight values ​​to the coordinate space of the capacity envelope domain; using the distribution of weight values ​​in the coordinate space as the efficiency gradient; and generating a time-series capacity map containing the capacity boundary and the efficiency gradient.

5. The virtual power plant trading and scheduling optimization system based on digital twins according to claim 1, characterized in that, The process by which the instruction twin verification module generates the target instruction sequence includes: reading the external interactive task package transmitted from the remote scheduling cloud, extracting the interactive quota containing the total task requirements, and reading the target regulation power value from the interactive quota; extracting the time window index containing the task execution period, and reading the task start and end timestamps from the time window index; obtaining the simulation time step parameter in the digital twin engine; using the simulation time step parameter to perform time-domain segmentation on the time range limited by the task start and end timestamps to generate discrete time nodes; mapping the target regulation power value to the discrete time nodes and generating instruction power values ​​corresponding to each discrete time node; and constructing the target instruction sequence based on the discrete time nodes and instruction power values.

6. The virtual power plant trading and scheduling optimization system based on digital twins according to claim 1, characterized in that, The process of the instruction twin verification module in identifying scheduling conflict periods includes: mapping discrete time nodes in the target instruction sequence to the time-series carrying capacity map and aligning them on the time axis; comparing the instruction power values ​​at each discrete time node with the upper limit threshold of the capacity envelope domain in the time-series carrying capacity map; marking discrete time nodes with instruction power values ​​greater than the upper limit threshold of the capacity envelope domain as scheduling conflict periods; calculating the difference between the instruction power values ​​and the upper limit threshold of the capacity envelope domain within the scheduling conflict period to obtain the overflow quota value; and marking discrete time nodes with instruction power values ​​less than the upper limit threshold of the capacity envelope domain as surplus capacity periods.

7. The virtual power plant trading and scheduling optimization system based on digital twins according to claim 1, characterized in that, The process of generating a revised scheduling task book by the instruction twin verification module includes: retrieving surplus capacity periods in the time-series load map based on the time axis index, where the time interval between scheduling conflict periods is within a preset threshold; calculating the difference between the upper limit threshold of the capacity envelope domain and the instruction power value of the surplus capacity period to obtain the allowable margin value; accumulating the overflow quota value into the instruction power value of the surplus capacity period, and performing numerical limiting processing on the accumulation result using the upper limit threshold of the capacity envelope domain; resetting the instruction power value of the scheduling conflict period to the upper limit threshold of the capacity envelope domain, and performing numerical truncation processing on the overflow quota value exceeding the allowable margin value; and integrating the instruction power values ​​of each discrete time node after resetting and accumulation processing to generate a revised scheduling task book.

8. The virtual power plant trading and scheduling optimization system based on digital twins according to claim 1, characterized in that, The process of generating a resource priority sequence by the capacity locking allocation module includes: extracting the device identification code of each virtual node in the digital twin space and reading the response gradient coefficient associated with the device identification code; using the numerical sorting logic built into the digital twin engine, performing descending sorting processing on the device identification codes according to the value of the response gradient coefficient; assigning a call order index to each device identification code according to the result of the descending sorting processing; and constructing a resource priority sequence based on the device identification code and the call order index.

9. A virtual power plant trading and scheduling optimization system based on digital twins according to claim 1, characterized in that, The process by which the capacity locking and allocation module generates and sends execution scheduling instructions to the terminal includes: reading the instruction power value and the call order index in the resource priority sequence from the modified scheduling task book; extracting the current available capacity of each virtual node according to the call order index; performing a step-by-step deduction calculation on the current available capacity of each virtual node using the instruction power value until the instruction power value is zero, and determining the actual task load of each virtual node; retrieving the physical network address associated with each virtual node; encapsulating the task load value and the physical network address to generate an execution scheduling instruction containing the execution action; and sending the execution scheduling instruction to the corresponding resource execution terminal through the communication interface.

Images