Intelligent power dispatching method and system for virtual power plant
By monitoring the status of distributed resource nodes in virtual power plants and improving the bee algorithm, combined with prediction deviation correction and communication reliability, a three-layer progressive optimization architecture is constructed to solve the problems of prediction error sensitivity and multi-time scale coordination in virtual power plant scheduling, and achieve efficient and stable power scheduling.
Patent Information
- Application Number
- CN202511214931.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-28
- Publication Date
- 2025-09-30
- Estimated Expiration
- 2045-08-28
AI Technical Summary
Existing virtual power plant scheduling methods lack an effective uncertainty handling mechanism, which makes the scheduling scheme sensitive to prediction errors, power imbalance, and lack of multi-time scale coordination mechanism, making it impossible to take into account both economic efficiency and rapid response needs.
By monitoring the status of distributed resource nodes in the virtual power plant, collecting real-time output, charge status and communication quality indicators, establishing an improved bee algorithm, introducing prediction bias correction terms and communication reliability weights, and constructing a three-layer progressive optimization architecture, multi-time scale coordination is achieved.
It improves the power prediction accuracy and the robustness of the scheduling scheme, ensures the economy and communication robustness of the scheduling scheme, achieves organic coordination between different time scales, and meets the needs of day-ahead planning and real-time control.
Smart Images

Figure CN120728758A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of power dispatching, and in particular to a method and system for intelligent power dispatching of a virtual power plant. Background Art
[0002] As an emerging distributed energy management model, virtual power plants (VPs) leverage information and communication technologies to aggregate geographically dispersed photovoltaic (PV) and wind power generation units, energy storage systems, and controllable load devices into a unified, dispatchable resource pool, enabling them to participate in electricity market transactions and grid operations. Existing VP scheduling methods primarily employ traditional centralized optimization algorithms, such as linear programming, quadratic programming, and particle swarm optimization. These methods establish economic scheduling mathematical models, minimize operating costs, and solve for the optimal scheduling scheme for each distributed resource, while satisfying power balance constraints and equipment output limits. These methods are effective in small-scale, deterministic environments and can achieve theoretically optimal solutions.
[0003] However, traditional optimization algorithms lack an effective uncertainty handling mechanism when dealing with the randomness and intermittency of renewable energy output and load demand, resulting in scheduling schemes being sensitive to prediction errors and frequent power imbalance problems in actual operation; secondly, existing methods do not fully consider the impact of differences in communication quality between distributed resource nodes on scheduling reliability. When some resources experience communication delays or interruptions, the entire scheduling system is prone to failure; thirdly, traditional algorithms lack multi-time scale coordination mechanisms and cannot simultaneously take into account the economic efficiency of day-ahead planning and the rapid response requirements of real-time control.
[0004] The existing technology lacks intelligent scheduling algorithms tailored to the distributed nature of virtual power plants. Specifically, key technical challenges remain, including building a distributed resource status awareness mechanism that considers communication reliability, establishing an intelligent optimization algorithm that incorporates forecast bias correction, and designing a progressive optimization architecture that coordinates across multiple timescales. Summary of the Invention
[0005] The present application provides an intelligent power dispatching method and system for a virtual power plant, which solves the state monitoring problem of lack of communication reliability awareness in the distributed resource dispatching of virtual power plants, the intelligent optimization dispatching problem of lack of adaptive correction of prediction deviations, and the progressive optimization architecture problem of lack of multi-time scale coordination.
[0006] In the first aspect, the present application provides an intelligent power dispatching method for a virtual power plant, which includes: performing status monitoring on each distributed resource node of the virtual power plant, collecting real-time output, charge status and communication quality indicators, and obtaining a resource status data set; performing power prediction based on the resource status data set, calculating the power prediction deviation in real time and triggering online correction to obtain a power prediction sequence; inputting the resource status data set and the power prediction sequence into an improved bee algorithm, and obtaining a target scheduling scheme for the virtual power plant through a neighborhood search including a prediction deviation correction term and a probabilistic selection based on communication reliability; establishing a three-layer progressive optimization architecture based on the target scheduling scheme, realizing multi-time scale coordination through time coupling constraints, and obtaining distributed resource power control instructions.
[0007] Optionally, the state monitoring of each distributed resource node of the virtual power plant is performed to collect real-time output, state of charge and communication quality indicators to obtain a resource state data set, including: The power measurement device deployed on the photovoltaic power generation unit collects the real-time power generation power and adjustable capacity range to obtain the photovoltaic power generation status parameters; A battery management system based on the energy storage system collects the state of charge, charge and discharge power limits, and cycle efficiency parameters to obtain energy storage operation state parameters; The intelligent controller of the controllable load device collects power demand, response time window and interruption allowable duration to obtain load response characteristic parameters; The photovoltaic power generation state parameters, the energy storage operation state parameters, the load response characteristic parameters and the communication quality indicators of each distributed resource node are standardized to obtain the resource state data set.
[0008] Optionally, performing power prediction according to the resource status data set, calculating the power prediction deviation in real time and triggering online correction to obtain a power prediction sequence includes: Inputting historical power generation data and real-time meteorological information in the resource status data set into a pre-trained time series neural network to perform power prediction and obtain a power prediction value; Calculate the difference between the power prediction value and the current actual power generation value to obtain the power prediction deviation; Comparing the absolute value of the power prediction deviation in three consecutive time windows with a threshold of 15% of the rated power, and triggering an online correction mechanism when the threshold is exceeded to obtain a correction trigger signal; Based on the correction trigger signal, a sliding window least square method is used to dynamically adjust and calculate the weight coefficient of the historical prediction data to obtain a corrected prediction parameter; The corrected prediction parameters are re-input into the pre-trained time series neural network to perform power prediction, thereby obtaining the power prediction sequence.
[0009] Optionally, the inputting the resource status dataset and the power prediction sequence into an improved honeybee algorithm, obtaining a target scheduling scheme for the virtual power plant through a neighborhood search including a prediction deviation correction term and a probabilistic selection based on communication reliability, includes: The distributed generation unit output and energy storage system charge state in the resource status dataset are used as initial values to initialize three types of agents: resource exploration agent, follow-up optimization agent, and global search agent, and obtain the improved bee algorithm population structure; The resource exploration agent performs a difference operation on the current scheduling position and the randomly selected position in the improved bee algorithm population structure to obtain a position offset; Calculate a neighborhood search correction term based on the position offset and the power prediction deviation in the power prediction sequence to obtain a new scheduling position including the prediction deviation correction; Performing economic cost calculation on the scheduling plan corresponding to the new scheduling position and weighting processing on the communication reliability weight of the resource status data to obtain a comprehensive fitness evaluation value; A probability selection wheel is constructed based on the comprehensive fitness evaluation value, and the follow-up optimization agent selects a high-quality scheduling solution according to the probability to perform local search optimization to obtain an improved scheduling solution; The improved scheduling scheme is verified by executing power balance equality constraints and inequality constraints on upper and lower limits of power generation unit output. When the global search agent detects a communication interruption, emergency reconstruction is performed to obtain the target scheduling scheme of the virtual power plant.
[0010] Optionally, the probability selection wheel is constructed based on the comprehensive fitness evaluation value, and the follow-up optimization agent selects a high-quality scheduling solution according to the probability to perform local search optimization to obtain an improved scheduling solution, including: Normalizing the comprehensive fitness evaluation value, mapping the fitness value of each scheduling scheme to the interval of 0 to 1, and obtaining a standardized fitness value; Calculating the cumulative probability distribution of each scheduling scheme according to the standardized fitness value, constructing a probability interval partitioning table, and obtaining the probability selection wheel; Generate a random probability value between 0 and 1 through a random number generator, perform interval matching search on the probability selection wheel, and obtain the selected target scheduling solution; Input the target scheduling plan into the follow-up optimization agent, execute a local search algorithm within its neighborhood to update the position, and obtain a locally optimized scheduling position; The economic cost and communication reliability weighted evaluation of the locally optimized scheduling position are recalculated, and after fitness comparison and screening, the improved scheduling solution is obtained.
[0011] Optionally, the step of inputting the target scheduling solution into the follow-up optimization agent, executing a local search algorithm within its neighborhood to update the position, and obtaining a locally optimized scheduling position includes: Calculating a neighborhood search radius according to the communication reliability weight of each distributed resource in the target scheduling scheme, and multiplying a basic search radius by the inverse of the communication reliability to obtain an adaptive neighborhood search range; Based on the adaptive neighborhood search range, a Gaussian random perturbation algorithm is used to perform position perturbations on the power generation unit output and energy storage charge state in the target scheduling scheme to obtain a set of candidate search positions; Calculating the power generation cost function and performing power balance constraint checks on the power configuration scheme of each position in the candidate search position set, screening the candidate positions that meet the constraint conditions, and obtaining feasible candidate positions; A fitness value comparison and evaluation is performed on the feasible candidate positions and the target scheduling solution, and the scheduling position with a better fitness value is selected as the update result to obtain the locally optimized scheduling position.
[0012] Optionally, establishing a three-layer progressive optimization architecture according to the target scheduling scheme, achieving multi-time scale coordination through time coupling constraints, and obtaining distributed resource power control instructions includes: Divide the target scheduling plan into three time scales: day-ahead scheduling layer, intraday scheduling layer, and real-time scheduling layer, and construct the three-layer progressive optimization architecture; Based on the three-layer progressive optimization architecture, time coupling constraints are set to limit the adjustment amount of the intraday scheduling layer to within 20% of the benchmark plan of the day-ahead scheduling layer, and the control instructions of the real-time scheduling layer are limited to the adjustment range of the intraday scheduling layer, thereby obtaining multi-time scale constraint parameters; Performing coordination checks on the scheduling plans of the day-ahead scheduling layer, the intraday scheduling layer, and the real-time scheduling layer according to the multi-time scale constraint parameters, calculating the scheduling deviation between the layers and performing constraint correction processing to obtain a coordinated hierarchical scheduling plan; The coordinated hierarchical scheduling scheme is sent to the local controller of each distributed resource node through an encrypted communication protocol, an execution status feedback monitoring mechanism is established, and the distributed resource power control instruction is obtained.
[0013] In a second aspect, the present application provides an intelligent power dispatching system for a virtual power plant, the intelligent power dispatching system for a virtual power plant comprising: The monitoring module is used to monitor the status of each distributed resource node in the virtual power plant, collect real-time output, charge status and communication quality indicators, and obtain a resource status data set; A correction module, configured to perform power prediction based on the resource status data set, calculate the power prediction deviation in real time and trigger online correction to obtain a power prediction sequence; A search module is configured to input the resource status data set and the power prediction sequence into an improved honeybee algorithm, and obtain a target scheduling plan for the virtual power plant through a neighborhood search including a prediction deviation correction term and a probabilistic selection based on communication reliability; The coordination module is used to establish a three-layer progressive optimization architecture according to the target scheduling scheme, realize multi-time scale coordination through time coupling constraints, and obtain distributed resource power control instructions.
[0014] In a third aspect, an intelligent power dispatching device for a virtual power plant is provided, comprising: a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor calls the instructions in the memory so that the intelligent power dispatching device for the virtual power plant executes the above-mentioned intelligent power dispatching method for the virtual power plant.
[0015] In a fourth aspect, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium, which, when executed on a computer, enables the computer to execute the above-mentioned intelligent power dispatching method of a virtual power plant.
[0016] In the technical solution provided in this application, by conducting comprehensive status monitoring of each distributed resource node of the virtual power plant, collecting real-time output, charge status and communication quality indicators, and obtaining a resource status data set including a communication reliability weight coefficient, it is possible to perceive the operating status and communication quality of each distributed resource in real time, effectively solving the scheduling failure problem caused by the traditional method ignoring the communication quality difference; by performing power prediction based on the resource status data set, calculating the power prediction deviation in real time and triggering the online correction mechanism, a dynamically corrected power prediction sequence is obtained, which can actively identify and compensate for the uncertainty of renewable energy output, significantly improve the power prediction accuracy and the robustness of the scheduling scheme, and overcome the limitation of the existing technology that is sensitive to prediction errors.
[0017] This application improves the honey bee algorithm by inputting resource status data sets and power prediction sequences. The innovation of this algorithm lies in the introduction of a neighborhood search mechanism including a prediction bias correction term and a probabilistic selection mechanism based on communication reliability. The prediction bias correction term enables the resource exploration agent to actively adjust the search direction according to the prediction error of renewable energy, and the introduction of the communication reliability weight enables the follow-up optimization agent to consider not only economic indicators but also the communication quality of each resource node when selecting a scheduling scheme. This dual optimization mechanism ensures that the scheduling scheme has both good economy and sufficient communication robustness, compared with the traditional honey bee algorithm's single mechanism that only relies on fitness value for selection. The improved algorithm of the present invention can better adapt to the special environment of virtual power plants that are distributed, heterogeneous, and have uncertain communications. By establishing a three-layer progressive optimization architecture of day-ahead scheduling layer, intraday scheduling layer, and real-time scheduling layer according to the target scheduling plan, and setting time coupling constraints to achieve multi-time scale coordination, distributed resource power control instructions are obtained. This architecture achieves organic coordination between different time scales by limiting the intraday adjustment amount to a specific proportion of the day-ahead benchmark plan and limiting the real-time control instructions to the intraday scheduling range, etc., which not only ensures the economic optimization of day-ahead scheduling, but also meets the rapid response requirements of intraday and real-time scheduling, effectively solving the problem that the existing technology lacks a multi-time scale coordination mechanism. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0019] Figure 1 A schematic diagram of an embodiment of a method for intelligent power dispatching of a virtual power plant in an embodiment of the present application; Figure 2 A schematic diagram of an embodiment of an intelligent power dispatching system for a virtual power plant in an embodiment of the present application; Figure 3 It is a schematic block diagram of the structure of the power intelligent dispatching equipment of the virtual power plant in an embodiment of the present invention. DETAILED DESCRIPTION
[0020] An embodiment of the present application provides a method and system for intelligent power dispatching of a virtual power plant. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or that are inherent to these processes, methods, products or devices.
[0021] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In one embodiment of the present application, an embodiment of the method for intelligent power dispatching of a virtual power plant includes: Step S101: Monitor the status of each distributed resource node of the virtual power plant, collect real-time output, state of charge, and communication quality indicators, and obtain a resource status data set; It is understandable that the execution subject of the present invention can be the power intelligent dispatching system of the virtual power plant, or it can be a terminal or a server, depending on the actual application requirements. The server is used as an example for explanation.
[0022] Specifically, the system acquires the operating status parameters of photovoltaic power generation units, wind power generation units, energy storage systems and controllable load equipment in real time through the intelligent data acquisition terminals deployed at each distributed resource node. The specific data collected includes: the real-time output of each distributed power generation unit , Adjustable capacity range , ramp rate constraint parameters; state of charge of the energy storage system , charge and discharge power limits, cycle efficiency parameters; key parameters such as the power demand of controllable loads, response time window, and allowed interruption duration.
[0023] In order to ensure high data reliability, the present invention introduces a distributed resource communication reliability assessment mechanism. Unlike the existing technology that only collects basic operation data, the present invention generates a communication reliability weight coefficient by calculating communication quality indicators such as packet loss rate, transmission delay time, signal strength, etc. of each resource node. This weight coefficient will directly affect the trust and call priority of the resource in subsequent scheduling decisions, thereby ensuring that the scheduling system can reasonably schedule various resources and improve the stability and efficiency of the overall operation.
[0024] In addition, in order to ensure data consistency and accuracy, the system has established a multi-source data fusion pre-processing module to standardize the collected heterogeneous data. and state of charge , the standardization process is expressed as:
[0025]
[0026] in, , are the minimum and maximum values of power, , is the minimum and maximum value of the state of charge. Through this standardization, the data can be unified into the range of [0,1], which is convenient for subsequent algorithm processing. By setting the threshold Perform anomaly detection on the data. For each collected data point , if its deviation exceeds the allowable range, such as ,in If the value is the historical average value of the resource, the data is considered abnormal and needs to be removed or corrected. Due to communication delays, the collected data may have a timestamp deviation. To compensate for this deviation, assume that the timestamp of each data point is , and the communication delay is , the adjusted timestamp is: Through compensation, the accuracy of data timestamps is ensured, preventing time synchronization issues from affecting scheduling decisions.
[0027] Step S102: Perform power prediction based on the resource status data set, calculate the power prediction deviation in real time and trigger online correction to obtain a power prediction sequence; Specifically, historical power generation data and real-time meteorological information are extracted from the resource status dataset and fed into a pre-trained time series neural network for power forecasting. This input data includes power generation data from the past 24 hours and real-time meteorological data such as wind speed, temperature, and humidity. This allows the model to account for the impact of environmental factors on power output and make appropriate forecasts.
[0028] The predicted power value is an approximation based on these input data. However, because actual generated power can be affected by various factors, there may be deviations between the predicted and actual values. Therefore, the system next calculates the power forecast deviation. The purpose of calculating the deviation is to quantify the model's prediction error and assess its performance in its current state. A large prediction deviation may indicate that the model needs to be adjusted to improve prediction accuracy. To ensure the accuracy of the prediction model, the system implements a threshold judgment mechanism. The system monitors the absolute value of the power forecast deviation within three consecutive time windows and compares it with a threshold of 15% of the rated power. This threshold judgment enables the system to identify cases with large deviations. If the deviation in any time window exceeds this threshold, the system triggers the online correction mechanism, generating a correction trigger signal.
[0029] The goal of the correction mechanism is to reduce future prediction errors by dynamically adjusting the model's prediction parameters. To this end, the system uses a sliding window least squares method to dynamically adjust and calculate the weight coefficients of historical prediction data. The sliding window method weights the data within a certain time range and uses the least squares method for optimization to update the weight of historical data in the prediction model. The core purpose of this process is to optimize the weight coefficients so that the model can better fit the changing trends of actual power generation and improve the accuracy of future predictions. Once the weight coefficients are updated, the system will re-input the revised prediction parameters into the pre-trained time series neural network and perform power prediction again. By using the revised prediction parameters, the model can better reflect the current system status and environmental changes, thereby generating a more accurate power prediction series.
[0030] Step S103: Input the resource status dataset and the power prediction sequence into the improved honeybee algorithm, and obtain the target scheduling plan of the virtual power plant through neighborhood search including a prediction deviation correction term and probabilistic selection based on communication reliability; Specifically, the system first obtains a resource status dataset and a power forecast sequence from steps S101 and S102. The resource status dataset contains information such as the real-time output, state of charge, and power demand of each distributed resource, while the power forecast sequence predicts power values based on historical data and meteorological information. This data is then input into the improved bee algorithm.
[0031] During the initialization phase of the improved bee algorithm, a set of bee positions is initialized, representing the current scheduling solution. Each position is initialized based on the current resource state and power forecast sequence. The bee positions are initialized to a combination of resource state and forecast values, such as the real-time output and power forecast values of photovoltaic power generation and energy storage systems.
[0032] During the exploration phase of the honey bee algorithm, the resource exploration agent determines the position offset by calculating the difference between its current position and a randomly selected position. It then uses the power prediction deviation to calculate a neighborhood search correction term. This correction term adjusts the coefficient to control the deviation strength, thereby updating the scheduled position to better align with the actual prediction deviation and improve the accuracy of the scheduling plan. The performance of the scheduling plan is evaluated based on economic cost and communication reliability. Communication reliability is weighted based on the communication quality of the resource node, such as packet loss rate and transmission delay. A comprehensive fitness evaluation value is calculated based on economic cost and communication reliability and used to probabilistically select a scheduling plan. The fitness values of all scheduling plans are normalized and mapped to the [0, 1] range. The cumulative probability distribution of the scheduling plans is then calculated, and a probability selection wheel is constructed to select the optimal scheduling plan using a random number generator.
[0033] During the local optimization phase, the follower optimization agent calculates an adaptive neighborhood search range based on the communication reliability weights of each resource node in the target scheduling solution. Poor communication quality results in a smaller search range to avoid over-searching unreliable resources. During this phase, a Gaussian random perturbation algorithm is used to perturb the power generation unit output and energy storage state of charge in the scheduling solution to generate new candidate solutions. For each candidate location, the power generation cost is calculated and the power balance constraint is verified. The power balance constraint ensures that the total generated power is balanced with the load demand, while also meeting the upper and lower output limits of the power generation units to ensure the feasibility of the generated scheduling solution in practical applications.
[0034] The innovation of step S103 lies in the introduction of multiple optimization mechanisms, particularly innovative methods such as neighborhood search correction terms, communication reliability weighting, and adaptive neighborhood search radius. Through these innovations, the system not only optimizes the scheduling plan but also adjusts the search strategy in real time based on communication quality, improving the scheduling accuracy and overall system stability.
[0035] Step S104: establishing a three-layer progressive optimization architecture according to the target scheduling scheme, achieving multi-time scale coordination through time coupling constraints, and obtaining distributed resource power control instructions.
[0036] Specifically, the target scheduling scheme is coordinated across timescales, and unified scheduling and control across multiple timescales is achieved by constructing a three-layer progressive optimization architecture. Although this step does not involve algorithmic innovation, it has important engineering significance and practical value in the actual operation of virtual power plants.
[0037] The target dispatch plan is divided into three tiers based on timescales: day-ahead dispatch, intraday dispatch, and real-time dispatch. This tiered approach provides a clear temporal structure for the dispatch plan, facilitating the development of long-term planning, mid-term adjustments, and short-term real-time control strategies, laying the foundation for flexible power system management.
[0038] On this basis, the system establishes time-coupling constraints to limit the range of adjustments between different levels. For example, adjustments to the previous day's baseline plan by the intraday scheduling layer are typically limited to 20%, while the operating range of the real-time scheduling layer is limited to the intraday scheduling adjustment range. This constraint relationship helps ensure scheduling stability, prevents sudden disruptions to the overall system by any one level, and improves responsiveness and coordination among multiple levels.
[0039] To further enhance the synergy between levels, the system performs consistency checks on the scheduling plans at each level, analyzes scheduling deviations, and performs necessary corrections accordingly. After adjustments, a coordinated hierarchical scheduling scheme is obtained, which not only meets physical constraints but also ensures scheduling consistency between levels, thus having higher practical operability and reliability. This coordinated scheduling scheme is sent to the local controller of each distributed resource node via encrypted communication, and a state feedback and monitoring mechanism is built to support it. This process ensures the secure transmission and closed-loop execution of scheduling instructions, ensuring that each resource node can operate according to plan, thereby achieving effective power control of distributed resources in the virtual power plant.
[0040] In a specific embodiment, the process of executing step S101 may specifically include the following steps: The power measurement device deployed on the photovoltaic power generation unit collects the real-time power generation power and adjustable capacity range to obtain the photovoltaic power generation status parameters; The battery management system based on the energy storage system collects the state of charge, charge and discharge power limit and cycle efficiency parameters to obtain the energy storage operation status parameters; The intelligent controller of the controllable load device collects power demand, response time window and interruption allowable duration to obtain load response characteristic parameters; The photovoltaic power generation state parameters, energy storage operation state parameters, load response characteristic parameters and communication quality indicators of each distributed resource node are standardized to obtain a resource state data set.
[0041] Specifically, power generation is monitored in real time using power measurement devices installed on photovoltaic power generation units. These measurement devices should accurately reflect the current output power of the photovoltaic power generation units and provide sufficient temporal resolution to accurately record changes in photovoltaic power generation capacity at all times. Furthermore, the measurement devices must provide adjustable capacity range parameters to dynamically adjust to varying environmental conditions and requirements. Sensors or systems monitor factors such as light intensity and ambient temperature to determine the power generation status of the photovoltaic system, ensuring accurate modeling of power generation capacity. The energy storage system's battery management system (BMS) monitors the battery's state of charge (SOC) and charge and discharge power limits in real time. For example, when the battery is fully charged, the system limits charging power to prevent overcharging. Conversely, when the battery charge is low, the system reduces discharge power to protect the battery and extend its life. Battery cycle efficiency is another important parameter, indicating the battery's ability to maintain efficient storage and release of energy during different charge and discharge cycles. This data can be used to analyze the health of the energy storage system and its role in the overall system. Load response characteristic parameters are primarily provided by intelligent load controllers connected to the system. The system needs to understand the load's power requirements and adjust the operating status of different load devices in real time through load scheduling algorithms. Furthermore, the system needs to collect the response time window and allowed interruption duration of each load device. This information will help optimize load scheduling and ensure system stability and reliability. Communication quality is crucial when communicating between multiple distributed resource nodes. To ensure reliable and real-time data transmission, the system needs to monitor communication quality indicators between each node, such as signal strength, transmission delay, and bit error rate. These communication quality indicators will help the system select the optimal communication path and scheduling strategy when optimizing load scheduling, power generation, and energy storage management, and avoid system efficiency reduction caused by poor communication. The collected photovoltaic power generation status parameters, energy storage system status parameters, load response characteristic parameters, and communication quality indicators need to be standardized. The purpose of standardization is to convert data from different sources into a unified metric, eliminating differences between data sources and facilitating subsequent analysis and decision-making. A specific standardization method can be to normalize each data item to ensure that each data item is compared within the same weight range. This approach prevents the characteristics of a particular data source from unnecessarily biasing the overall decision-making process.
[0042] In a specific embodiment, the process of executing step S102 may specifically include the following steps: The historical power generation data and real-time meteorological information in the resource status dataset are input into the pre-trained time series neural network to perform power forecasting and obtain the power forecast value; The power forecast deviation is obtained by calculating the difference between the power forecast value and the current actual power generation value; The absolute value of the power prediction deviation in three consecutive time windows is compared with a threshold of 15% of the rated power. When the threshold is exceeded, an online correction mechanism is triggered to obtain a correction trigger signal. Based on the correction trigger signal, the sliding window least square method is used to dynamically adjust the weight coefficient of the historical prediction data to obtain the corrected prediction parameters; The corrected prediction parameters are re-input into the pre-trained time series neural network for power prediction to obtain a power prediction sequence.
[0043] Specifically, the system inputs actual power generation over a period of time from resource status data, along with temperature, humidity, and light intensity, into a pre-trained time series neural network. This neural network, after sufficient training, can predict future power generation based on historical data and meteorological conditions. During the input process, the system ensures the temporal synchronization and accuracy of the data so that the neural network can extract underlying patterns and make accurate predictions. Next, the system compares the power prediction output by the neural network with the current actual power generation value, calculating the difference between the two to obtain the power prediction deviation. This deviation represents the error between the prediction and the actual situation. To ensure prediction accuracy, the system may set a certain tolerance range. If the deviation fluctuates within a certain range, the prediction is considered accurate. If the deviation exceeds the set tolerance range, a correction process begins. The system monitors changes in the power prediction deviation, specifically the absolute value of the power prediction deviation within three consecutive time windows. If the deviation exceeds the 15% error threshold for each of the three consecutive time windows, the model is considered to have a significant error and requires correction. If the set threshold is exceeded, the system automatically triggers the online correction mechanism, generating a correction trigger signal to initiate the correction process. Based on the correction trigger signal, the system will use the sliding window least squares method to dynamically adjust the weight coefficients of the historical forecast data. Specifically, the sliding window method will select the latest forecast data as a reference over a period of time, and gradually adjust the weight of the historical data in the neural network, so that the recent data has a greater impact and the impact of the long-term data gradually decreases. Through the least squares optimization algorithm, the system will calculate a new set of weight coefficients to minimize the error, ensuring that in future power forecasts, the contribution of historical data is more accurate and the revised forecast results are more in line with the actual situation. Using the dynamically adjusted weight coefficients, the system will re-input the revised forecast parameters into the pre-trained time series neural network. The revised neural network will re-forecast the power and generate an updated power forecast sequence. This new power forecast sequence will be more accurate than the previous forecast and can better reflect the trend of actual power generation.
[0044] In a specific embodiment, the process of executing step S103 may specifically include the following steps: The output of distributed generation units and the state of charge of energy storage systems in the resource status dataset are used as initial values to initialize three types of agents: resource exploration agent, follow-up optimization agent, and global search agent, and the improved bee algorithm population structure is obtained. The resource exploration agent performs a difference operation between the current scheduling position and the randomly selected position in the improved bee algorithm population structure to obtain the position offset; Calculate the neighborhood search correction term based on the position offset and the power prediction deviation in the power prediction sequence to obtain a new scheduling position including the prediction deviation correction; The scheduling plan corresponding to the new scheduling position is subjected to economic cost calculation and weighted processing of communication reliability weight in resource status data to obtain a comprehensive fitness evaluation value; Based on the comprehensive fitness evaluation value, a probability selection wheel is constructed, and the optimization agent selects the best scheduling solution according to the probability to perform local search optimization and obtain the improved scheduling solution. The improved scheduling scheme is verified by executing power balance equality constraints and inequality constraints on the upper and lower limits of power generation unit output. When the global search agent detects communication interruption, emergency reconstruction is performed to obtain the target scheduling scheme of the virtual power plant.
[0045] Specifically, the system first extracts output data for distributed generation units and the state of charge of the energy storage system from the resource status dataset. These data serve as initial values, reflecting the current system's energy production capacity and storage status. Based on this initial data, the system initializes three types of agents: a resource exploration agent, a follow-up optimization agent, and a global search agent. Each agent plays a different role in the algorithm. The resource exploration agent is responsible for extensively exploring the possible scheduling space and searching for potential optimal solutions. The follow-up optimization agent focuses on optimizing the currently preferred scheduling solution locally and making refinement adjustments. The global search agent is responsible for searching for scheduling solutions from a global perspective, ensuring that the algorithm does not become trapped in local optimal solutions. The collaborative work of these agents forms a diverse population structure of the improved honey bee algorithm, enabling efficient optimization within the search space. The resource exploration agent uses a difference operation to compare the current scheduling position with a randomly selected position, generating a position offset. This offset represents the difference between the current scheduling solution and the new solution explored. In this way, the resource exploration agent continuously searches for new possible scheduling positions, increasing the breadth of its search. Based on the calculated position offset, the system combines the prediction deviation in the power forecast sequence to perform a neighborhood search correction. This allows the system to adjust the dispatch position based on the offset, thereby correcting for the impact of prediction errors. This correction process ensures that the dispatch position is closer to actual operational demand and takes into account the impact of prediction deviation on the dispatch plan, avoiding large errors. For the dispatch plan corresponding to the new dispatch position, the system performs an economic cost calculation to assess the cost-effectiveness of the plan. This typically includes the costs of power generation, energy storage, and load scheduling, as well as costs associated with energy conversion and transmission. The system also weights the dispatch plan based on the communication reliability indicator in the resource status data. This means that resource nodes with poor communication quality are appropriately weighted to avoid scheduling errors caused by unstable communication. After weighting, the system generates a comprehensive fitness evaluation value, which is used to evaluate the overall performance of the dispatch plan. Based on this comprehensive fitness evaluation value, the system constructs a probabilistic selection wheel, where each dispatch plan has a different probability of being selected based on its fitness value. Dispatches with higher fitness have a greater chance of being selected for subsequent optimization. The follow-up optimization agent uses a selection wheel to probabilistically select the best scheduling solution and conducts local search optimization around it. Through this local optimization, the system can further improve the accuracy of the scheduling solution and refine the currently selected solution. Within the optimized scheduling solution, the system needs to perform several constraint verifications to ensure that the scheduling solution meets actual operational requirements. The first is the power balance equality constraint, which requires that the difference between the power output of all generation units and energy storage units and the load demand must be zero to ensure power balance in the grid.Secondly, the system also needs to verify the inequality constraints on the upper and lower limits of the power generation unit output to ensure that the output of each distributed power generation unit does not exceed its maximum output or fall below its minimum output, thereby avoiding excessive load or resource waste. During system operation, the global search agent continuously monitors the system status, paying particular attention to communication quality and node operation. If an abnormality such as communication interruption or resource node failure is detected, the system immediately activates the emergency reconstruction mechanism and quickly adjusts the scheduling plan to respond to the emergency. After optimization, correction, and verification of the above steps, the system ultimately generates the target scheduling plan for the virtual power plant. This plan can optimally balance various resources such as power generation, energy storage, load, and communication, and can respond to possible emergencies to ensure the stable operation of the power system.
[0046] In a specific embodiment, the execution step constructs a probability selection wheel based on the comprehensive fitness evaluation value, and the process of the following optimization agent selecting a high-quality scheduling solution according to the probability to perform local search optimization can specifically include the following steps: Perform numerical normalization on the comprehensive fitness evaluation value, map the fitness value of each scheduling scheme to the range of 0 to 1, and obtain the standardized fitness value; The cumulative probability distribution of each scheduling scheme is calculated based on the standardized fitness value, and a probability interval partition table is constructed to obtain a probability selection roulette wheel; A random number generator generates a random probability value between 0 and 1, and performs interval matching search in the probability selection wheel to obtain the selected target scheduling plan; The target scheduling plan is input into the follower optimization agent, and the local search algorithm is executed within its neighborhood to update the position and obtain the locally optimized scheduling position; The economic cost and communication reliability weighted evaluation of the locally optimized scheduling position are recalculated, and the improved scheduling plan is obtained after fitness comparison and screening.
[0047] Specifically, to ensure fair comparison and selection of the fitness values of different scheduling solutions, the comprehensive fitness evaluation values of all scheduling solutions are normalized. This normalization process maps all fitness values to a uniform range of 0 to 1. The normalized fitness values represent the relative performance of each scheduling solution relative to the optimal solution, facilitating the subsequent probabilistic selection process. The normalized fitness values are converted into a cumulative probability distribution for each scheduling solution. The cumulative probability represents the ratio of each scheduling solution's fitness to the sum of all solution fitnesses. Based on these cumulative probabilities, a probability interval partitioning table is constructed, assigning each scheduling solution an interval. The size of these intervals is proportional to the fitness of each solution, with solutions with higher fitness occupying larger intervals. In the probability selection roulette wheel, a random number generator generates a random probability value between 0 and 1. This value is used to search for a matching interval in the probability roulette wheel. The selected scheduling solution is determined by comparing the randomly generated probability value with the cumulative probabilities of the intervals in the roulette wheel. For example, if the random value falls within a certain interval, the scheduling solution corresponding to that interval is selected. This selection method ensures that scheduling plans with higher fitness are more likely to be selected. Once a target scheduling plan is selected, the follower optimization agent inputs the target scheduling plan into its own optimization framework and executes a local search algorithm within the neighborhood of the selected scheduling plan. The local search algorithm refines and optimizes the current plan by adjusting certain scheduling plan parameters or locations, such as power generation unit output and energy storage status. Local search algorithms can employ a variety of methods, such as gradient descent, simulated annealing, and local neighborhood search, depending on the system's requirements and complexity. The goal of local search optimization is to improve the accuracy of the scheduling plan and achieve optimal performance. The locally optimized scheduling plan undergoes another economic cost calculation and communication reliability weighted evaluation. The economic cost calculation is used to assess the capital expenditure required to implement the plan, including power generation costs, energy storage costs, and load scheduling-related expenses. Furthermore, the system weights the scheduling plans based on the communication quality of the resource status data. If the communication quality of certain resource nodes is poor, their scheduling weight is appropriately reduced to minimize the negative impact of communication failures or delays on system scheduling. After comprehensively weighting economic cost and communication reliability, the final fitness evaluation of the scheme is obtained. Among all locally optimized scheduling schemes, the optimal one is selected through fitness comparison. A scheme with a higher fitness indicates better performance in terms of both economic efficiency and communication reliability. After fitness comparison, the system selects the optimal improved scheduling scheme, which becomes the final scheduling scheme for the current scheduling cycle.
[0048] In a specific embodiment, the process of inputting the target scheduling solution into the follow-up optimization agent may specifically include the following steps: The neighborhood search radius is calculated based on the communication reliability weight of each distributed resource in the target scheduling scheme, and the adaptive neighborhood search range is obtained by multiplying the basic search radius by the inverse of the communication reliability. Based on the adaptive neighborhood search range, the Gaussian random perturbation algorithm is used to perturb the power generation unit output and energy storage charge state in the target scheduling scheme to obtain a set of candidate search locations. The power configuration scheme of each location in the candidate search location set is subjected to power generation cost function calculation and power balance constraint test, and the candidate locations that meet the constraint conditions are screened to obtain feasible candidate locations; The fitness values of the feasible candidate positions and the target scheduling scheme are compared and evaluated, and the scheduling position with a better fitness value is selected as the update result to obtain the locally optimized scheduling position.
[0049] Specifically, the communication reliability weight refers to the degree to which the communication quality of each distributed resource affects system scheduling within the scheduling plan. Based on these weights, the system first calculates a neighborhood search radius for each resource. The neighborhood search radius reflects the range that can be adjusted during local optimization. Resource nodes with poor communication quality require a smaller neighborhood search radius, while resource nodes with better communication quality can have a larger neighborhood search range. After obtaining the adaptive neighborhood search range, the system uses a Gaussian random perturbation algorithm to perturb the positions of key parameters in the target scheduling plan (such as the output of the power generation unit and the state of charge of the energy storage system). The Gaussian perturbation algorithm randomly generates perturbation values that conform to a normal distribution and adds these perturbations to the current scheduling position, generating multiple possible candidate search locations. This Gaussian perturbation ensures that the perturbation amplitude remains within the system's defined neighborhood search range, ensuring that the optimization process is neither overly aggressive nor misses potential high-quality solutions. In this way, the system can search for different scheduling solutions within the local area and evaluate their performance. For the generated set of candidate search locations, the system first calculates the corresponding power configuration for each candidate location. This includes the power output of all generating units and the state of charge of the energy storage system. The system then calculates the total cost of each candidate solution using a power generation cost function. Power generation costs typically include fuel costs, energy conversion fees for charging and discharging energy storage, and other operating and maintenance costs. Furthermore, the system verifies the power balance constraints of each candidate solution, ensuring that the combined output of all generating units and energy storage equipment can balance the load demand and that the output of each device does not exceed its upper limit or fall below its lower limit. Solutions that meet these constraints are considered "feasible candidate locations." The system compares and evaluates all feasible candidate locations based on their fitness values. Fitness values typically include factors such as the cost of the scheduling solution, the balance between generation and storage, and communication quality. Solutions with higher fitness values excel in multiple aspects, such as lower cost, better power balance, and more reliable communication. After comparison, the system selects the scheduling solution with the better fitness value as the local optimization result. This optimization process, based on the optimization objective of the target scheduling solution, continuously adjusts and selects the optimal solution, resulting in a final scheduling solution that achieves higher overall efficiency. Ultimately, after the aforementioned calculations, screening, and comparisons, the system outputs a locally optimized dispatch location. This location represents the optimal dispatch solution within the current neighborhood after perturbation and optimization. It effectively balances resources, meets the system's dispatch objectives, and maximizes both economic efficiency and communication reliability.
[0050] In a specific embodiment, the process of executing step S104 may specifically include the following steps: The target scheduling plan is divided into three time scales: day-ahead scheduling, intraday scheduling, and real-time scheduling, to build a three-layer progressive optimization architecture. Based on a three-layer progressive optimization architecture, time coupling constraints are set to limit the adjustment of the intraday scheduling layer to within 20% of the baseline plan of the day-ahead scheduling layer. The control instructions of the real-time scheduling layer are limited to the adjustment range of the intraday scheduling layer, thus obtaining multi-time scale constraint parameters. The coordination of the scheduling plans of the day-ahead scheduling layer, the intraday scheduling layer, and the real-time scheduling layer is checked based on multi-time scale constraint parameters. The scheduling deviation between the layers is calculated and constraint correction processing is performed to obtain a coordinated hierarchical scheduling plan. The coordinated hierarchical scheduling scheme is sent to the local controller of each distributed resource node through an encrypted communication protocol, and an execution status feedback monitoring mechanism is established to obtain distributed resource power control instructions.
[0051] Specifically, to address scheduling needs at different time scales, the target scheduling plan is first divided into three time scales: day-ahead scheduling, intraday scheduling, and real-time scheduling. Each level has different scheduling objectives and timeliness requirements. The day-ahead scheduling level is primarily responsible for developing an overall power generation and load plan based on long-term forecast data, providing a baseline plan for intraday and real-time scheduling. Based on the day-ahead scheduling plan, the intraday scheduling level makes adjustments based on real-time data (such as changes in power demand and fluctuations in renewable energy generation) to address dynamic intraday changes. The real-time scheduling level is responsible for adjusting power output based on current instantaneous data and real-time feedback, ensuring that the system can accurately respond to demand changes and promptly address emergencies during actual operation. These three levels work together through progressive optimization over time to achieve optimal scheduling results within different time scales. Within this three-level progressive optimization architecture, temporal coupling constraints are required between scheduling plans at each level to ensure coordination among the levels. Adjustments to the intraday scheduling level are limited to within 20% of the baseline plan at the day-ahead scheduling level. In other words, adjustments to the intraday scheduling plan must be limited to avoid deviations from the original general direction. This ensures system stability and avoids frequent, drastic adjustments. Control instructions at the real-time scheduling layer are constrained within the adjustment range of the intraday scheduling layer. This means that real-time scheduling must operate within the adjustment range of the intraday scheduling layer, ensuring that real-time instructions do not deviate from the established scheduling framework, thereby avoiding over-adjustment. These constraints help ensure that scheduling plans across all layers are coordinated and consistent, avoiding inconsistent scheduling plans across multiple timescales. During execution, the system verifies the coordination of the scheduling plans at the day-ahead, intraday, and real-time scheduling layers. The revised scheduling plan is more coordinated, ensuring that the scheduling plans at each layer are interconnected and achieve global optimization goals across multiple timescales. Once the coordinated hierarchical scheduling plan is finalized, the system sends it to the local controller of each distributed resource node via an encrypted communication protocol. Encrypted communication ensures the security and reliability of scheduling data during transmission, preventing data tampering or leakage. When the local controller of each resource node executes scheduling instructions, the system establishes an execution status feedback monitoring mechanism. Each distributed resource node feeds back its execution status to the central system, including information such as power output, equipment operating status, and control instruction execution. Through a feedback monitoring mechanism, the system can track the effectiveness of scheduling execution in real time and make adjustments when necessary. Based on the coordinated scheduling plan and feedback monitoring results, the system generates final distributed resource power control instructions. These instructions instruct each resource node (such as photovoltaic power generation units, energy storage systems, and load devices) to adjust its power output according to the established plan to ensure power balance and resource scheduling objectives across the virtual power plant. By executing these control instructions, the virtual power plant can respond to different demands and changes in real time, ensuring the efficient and stable operation of the power system.
[0052] The above describes the power intelligent dispatching method of the virtual power plant in the embodiment of the present application. The following describes the power intelligent dispatching system of the virtual power plant in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the power intelligent dispatching system of the virtual power plant includes: Monitoring module 201 is used to monitor the status of each distributed resource node of the virtual power plant, collect real-time output, charge state and communication quality indicators, and obtain a resource status data set; A correction module 202 is configured to perform power prediction based on the resource status data set, calculate power prediction deviation in real time, and trigger online correction to obtain a power prediction sequence; A search module 203 is configured to input the resource status dataset and the power prediction sequence into an improved honeybee algorithm, and obtain a target scheduling solution for the virtual power plant through a neighborhood search including a prediction deviation correction term and a probabilistic selection based on communication reliability; The coordination module 204 is configured to establish a three-layer progressive optimization architecture according to the target scheduling scheme, implement multi-time scale coordination through time coupling constraints, and obtain distributed resource power control instructions.
[0053] above Figure 2 The intelligent power dispatching system of the virtual power plant in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The intelligent power dispatching equipment of the virtual power plant in the embodiment of the present invention is described in detail from the perspective of hardware processing.
[0054] Reference Figure 3 In the embodiment of the present invention, a power intelligent dispatching device for a virtual power plant is also provided. The power intelligent dispatching device for the virtual power plant can be a server, and its internal structure can be as follows: Figure 3 As shown. The power intelligent dispatching equipment of the virtual power plant includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the computer-designed processor is used to provide computing and control capabilities. The memory of the power intelligent dispatching equipment of the virtual power plant includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the power intelligent dispatching equipment of the virtual power plant is used to store the corresponding data in this embodiment. The network interface of the power intelligent dispatching equipment of the virtual power plant is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.
[0055] Those skilled in the art will understand that Figure 3The structure shown in is only a block diagram of part of the structure related to the solution of the present invention, and does not constitute a limitation on the power intelligent dispatching equipment of the virtual power plant to which the solution of the present invention is applied.
[0056] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the intelligent power dispatching method of the virtual power plant.
[0057] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0058] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a power intelligent dispatching device of a virtual power plant (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, and other media that can store program code.
[0059] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.
Claims
1. A method for intelligent power dispatching of a virtual power plant, characterized in that: The method comprises: Monitor the status of each distributed resource node in the virtual power plant, collect real-time output, state of charge, and communication quality indicators, and obtain a resource status data set; Perform power prediction based on the resource status data set, calculate the power prediction deviation in real time and trigger online correction to obtain a power prediction sequence; Inputting the resource status data set and the power prediction sequence into the improved honeybee algorithm, obtaining the target scheduling plan of the virtual power plant through neighborhood search including a prediction deviation correction term and probabilistic selection based on communication reliability; A three-layer progressive optimization architecture is established according to the target scheduling scheme, and multi-time scale coordination is achieved through time coupling constraints to obtain distributed resource power control instructions.
2. The method for intelligent power dispatching of a virtual power plant according to claim 1, characterized in that: The state monitoring of each distributed resource node of the virtual power plant is performed, and real-time output, charge state and communication quality indicators are collected to obtain a resource state data set, including: The power measurement device deployed on the photovoltaic power generation unit collects the real-time power generation power and adjustable capacity range to obtain the photovoltaic power generation status parameters; A battery management system based on the energy storage system collects the state of charge, charge and discharge power limits, and cycle efficiency parameters to obtain energy storage operation state parameters; The intelligent controller of the controllable load device collects power demand, response time window and interruption allowable duration to obtain load response characteristic parameters; The photovoltaic power generation state parameters, the energy storage operation state parameters, the load response characteristic parameters and the communication quality indicators of each distributed resource node are standardized to obtain the resource state data set.
3. The method for intelligent power dispatching of a virtual power plant according to claim 1, characterized in that: The power prediction is performed according to the resource status data set, the power prediction deviation is calculated in real time and an online correction is triggered to obtain a power prediction sequence, including: Inputting historical power generation data and real-time meteorological information in the resource status data set into a pre-trained time series neural network to perform power prediction and obtain a power prediction value; Calculate the difference between the power prediction value and the current actual power generation value to obtain the power prediction deviation; Comparing the absolute value of the power prediction deviation in three consecutive time windows with a threshold of 15% of the rated power, and triggering an online correction mechanism when the threshold is exceeded to obtain a correction trigger signal; Based on the correction trigger signal, a sliding window least square method is used to dynamically adjust and calculate the weight coefficient of the historical prediction data to obtain a corrected prediction parameter; The corrected prediction parameters are re-input into the pre-trained time series neural network to perform power prediction, thereby obtaining the power prediction sequence.
4. The method for intelligent power dispatching of a virtual power plant according to claim 1, characterized in that: The resource status data set and the power prediction sequence are input into the improved honeybee algorithm, and a target scheduling scheme of the virtual power plant is obtained through a neighborhood search including a prediction deviation correction term and a probabilistic selection based on communication reliability, including: The distributed generation unit output and energy storage system charge state in the resource status dataset are used as initial values to initialize three types of agents: resource exploration agent, follow-up optimization agent, and global search agent, and obtain the improved bee algorithm population structure; The resource exploration agent performs a difference operation on the current scheduling position and the randomly selected position in the improved bee algorithm population structure to obtain a position offset; Calculate a neighborhood search correction term based on the position offset and the power prediction deviation in the power prediction sequence to obtain a new scheduling position including the prediction deviation correction; Performing economic cost calculation on the scheduling plan corresponding to the new scheduling position and weighting processing on the communication reliability weight of the resource status data to obtain a comprehensive fitness evaluation value; A probability selection wheel is constructed based on the comprehensive fitness evaluation value, and the follow-up optimization agent selects a high-quality scheduling solution according to the probability to perform local search optimization to obtain an improved scheduling solution; The improved scheduling scheme is verified by executing power balance equality constraints and inequality constraints on upper and lower limits of power generation unit output. When the global search agent detects a communication interruption, emergency reconstruction is performed to obtain the target scheduling scheme of the virtual power plant.
5. The method for intelligent power dispatching of a virtual power plant according to claim 4, characterized in that: The probability selection wheel is constructed based on the comprehensive fitness evaluation value, and the follow-up optimization agent selects a high-quality scheduling solution according to the probability to perform local search optimization to obtain an improved scheduling solution, including: Normalizing the comprehensive fitness evaluation value, mapping the fitness value of each scheduling scheme to the interval of 0 to 1, and obtaining a standardized fitness value; Calculating the cumulative probability distribution of each scheduling scheme according to the standardized fitness value, constructing a probability interval partitioning table, and obtaining the probability selection wheel; Generate a random probability value between 0 and 1 through a random number generator, perform interval matching search on the probability selection wheel, and obtain the selected target scheduling solution; Input the target scheduling plan into the follow-up optimization agent, execute a local search algorithm within its neighborhood to update the position, and obtain a locally optimized scheduling position; The economic cost and communication reliability weighted evaluation of the locally optimized scheduling position are recalculated, and after fitness comparison and screening, the improved scheduling solution is obtained.
6. The method for intelligent power dispatching of a virtual power plant according to claim 5, characterized in that: The target scheduling scheme is input into the follow-up optimization agent, and a local search algorithm is executed within its neighborhood to update the position to obtain a locally optimized scheduling position, including: Calculating a neighborhood search radius according to the communication reliability weight of each distributed resource in the target scheduling scheme, and multiplying a basic search radius by the inverse of the communication reliability to obtain an adaptive neighborhood search range; Based on the adaptive neighborhood search range, a Gaussian random perturbation algorithm is used to perform position perturbations on the power generation unit output and energy storage charge state in the target scheduling scheme to obtain a set of candidate search positions; Calculating the power generation cost function and performing power balance constraint checks on the power configuration scheme of each position in the candidate search position set, screening the candidate positions that meet the constraint conditions, and obtaining feasible candidate positions; A fitness value comparison and evaluation is performed on the feasible candidate positions and the target scheduling solution, and the scheduling position with a better fitness value is selected as the update result to obtain the locally optimized scheduling position.
7. The method for intelligent power dispatching of a virtual power plant according to claim 1, characterized in that: The method of establishing a three-layer progressive optimization architecture according to the target scheduling scheme, achieving multi-time scale coordination through time coupling constraints, and obtaining distributed resource power control instructions includes: Divide the target scheduling plan into three time scales: day-ahead scheduling layer, intraday scheduling layer, and real-time scheduling layer, and construct the three-layer progressive optimization architecture; Based on the three-layer progressive optimization architecture, time coupling constraints are set to limit the adjustment amount of the intraday scheduling layer to within 20% of the benchmark plan of the day-ahead scheduling layer, and the control instructions of the real-time scheduling layer are limited to the adjustment range of the intraday scheduling layer, thereby obtaining multi-time scale constraint parameters; Performing coordination checks on the scheduling plans of the day-ahead scheduling layer, the intraday scheduling layer, and the real-time scheduling layer according to the multi-time scale constraint parameters, calculating the scheduling deviation between the layers and performing constraint correction processing to obtain a coordinated hierarchical scheduling plan; The coordinated hierarchical scheduling scheme is sent to the local controller of each distributed resource node through an encrypted communication protocol, an execution status feedback monitoring mechanism is established, and the distributed resource power control instruction is obtained.
8. An intelligent power dispatching system for a virtual power plant, characterized in that: A method for implementing intelligent power dispatching of a virtual power plant according to any one of claims 1 to 7, wherein the intelligent power dispatching system of the virtual power plant comprises: The monitoring module is used to monitor the status of each distributed resource node in the virtual power plant, collect real-time output, charge status and communication quality indicators, and obtain a resource status data set; A correction module, configured to perform power prediction based on the resource status data set, calculate the power prediction deviation in real time and trigger online correction to obtain a power prediction sequence; A search module is configured to input the resource status data set and the power prediction sequence into an improved honeybee algorithm, and obtain a target scheduling plan for the virtual power plant through a neighborhood search including a prediction deviation correction term and a probabilistic selection based on communication reliability; The coordination module is used to establish a three-layer progressive optimization architecture according to the target scheduling scheme, realize multi-time scale coordination through time coupling constraints, and obtain distributed resource power control instructions.
9. An intelligent power dispatching device for a virtual power plant, characterized in that: It includes a memory and a processor, the memory stores a computer program that can be run on the processor, and the processor implements the power intelligent dispatching method of the virtual power plant described in any one of claims 1 to 7 when executing the computer program.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the processor is enabled to perform the method for intelligent power dispatching of a virtual power plant according to any one of claims 1 to 7.
Citation Information
Patent Citations
Virtual power plant cluster resource replacement method and device, electronic equipment and medium
CN118761575A
Power system optimization method based on electricity-carbon collaborative virtual power plant
CN119784069A
Energy scheduling method, device and equipment based on virtual power plant and storage medium
CN119994888A
Virtual aggregation system and method for regional energy source complex
WO2021244000A1
Online scheduling method and apparatus for wind-storage virtual power plant
WO2024067521A2
Cited By
Hierarchical collaborative intelligent scheduling method and system for virtual power plant based on multi-objective optimization
CN121395579A
A hierarchical collaborative intelligent scheduling method and system for virtual power plants based on multi-objective optimization
CN121395579B
Power distribution network voltage control method based on reactive virtual power plant
CN121791344A
A power distribution network voltage control method based on reactive virtual power plant
CN121791344B
Virtual power plant response scheduling method and device based on dynamic Bayesian game excitation
CN122026522A
