Virtual power plant resource scheduling method and device

Abstract

The invention discloses a virtual power plant resource scheduling method and device, and relates to the technical field of power grids. The method comprises the following steps: updating variable coefficients of a load side resource model and a dynamic response characteristic model of each heterogeneous resource of a virtual power plant according to current operation data of each heterogeneous resource collected in real time; performing analysis and aggregation processing on variable parameters and variable relations of the load side resource model and the dynamic response characteristic model of each heterogeneous resource after the variable coefficients are updated to obtain a multi-time-sequence aggregation scheduling model; processing based on the multivariate uncertainty index, and determining a feasible region in each scheduling period of the virtual power plant; adopting a multi-modal depth model to obtain a load prediction curve and an uncertainty interval of the current scheduling period; and scheduling each heterogeneous resource based on the feasible region, the load prediction curve and the uncertainty interval.

Description

Technical Field

[0001] This application relates to the field of power grid technology, and in particular to a method and apparatus for virtual power plant resource scheduling. Background Technology

[0002] With the deepening of energy transition and the continuous reform of the power market, Virtual Power Plants (VPPs), as an emerging technology that aggregates distributed energy resources, energy storage systems, and load-side adjustable resources to participate in grid operation and market transactions, have received widespread attention. Their core value lies in "aggregating" massive, dispersed, and heterogeneous load-side resources, providing the grid with flexible regulation capabilities and improving the economic benefits for resource owners. Currently, to rationally schedule and regulate VPP resources, fixed response characteristic models (such as adjustable capacity, response time, and duration) and resource models are typically established for various load resources (such as industrial equipment, electric vehicle charging piles, and commercial building air conditioning) based on offline identification or historical data. Fixed feasible regions and defined load operation trend curves are used as inputs to each response characteristic model and resource model, and scheduling is then performed on various heterogeneous resources based on the output results. The operating environment of virtual power plants is full of uncertainties, and the actual operating status of their heterogeneous resources is also constantly changing. Therefore, the static scheduling method based on the static response characteristic model, which does not consider the impact of changes in environment and operating status, does not match the actual scheduling capacity, resulting in low efficiency of resource scheduling and control of virtual power plants. Summary of the Invention

[0003] This invention provides a virtual power plant resource scheduling method and apparatus to at least solve the problem of low efficiency in virtual resource scheduling in related technologies. The technical solution of this invention is as follows: According to a first aspect of the present invention, a virtual power plant resource scheduling method is provided. The method includes: updating the variable coefficients of the load-side resource model and the variable coefficients of the dynamic response characteristic model of each heterogeneous resource in the virtual power plant based on real-time collected current operating data of each heterogeneous resource; the load-side resource model includes a representation of the relationship between the resource operating status of each heterogeneous resource and its corresponding physical characteristic indicators and resource environment; the dynamic response model includes a representation of the dynamic relationship between the economic response indicators of the heterogeneous resource and user behavior economic indicators and the electricity market environment; according to multiple time scales, analyzing, standardizing, and aggregating the variable parameters and variable relationships of each heterogeneous resource at each time scale in the updated load-side resource model and dynamic response characteristic model to obtain a multi-time-series aggregated scheduling model; and based on the split-bar optimization algorithm, using the minimum revenue index... To achieve this goal, we process multiple uncertainty indicators to obtain the feasible domain for each scheduling cycle of the virtual power plant. These multiple uncertainty indicators include renewable energy output, load fluctuations, and market electricity prices. A multimodal deep model is used to predict the load trend of the total load of the virtual power plant, represented by multi-source data, resulting in the load forecast curve and uncertainty interval for the current scheduling cycle. The multi-source data includes load data, meteorological data, time-specific data, traffic flow data, and event calendar data. The multimodal deep model is a reinforcement learning model. Based on the feasible domain, load forecast curve, and uncertainty interval, different inputs to a multi-time-series aggregated scheduling model are generated, and the output results of the multi-time-series aggregated scheduling model corresponding to different inputs are evaluated. The resource index parameters of each heterogeneous resource corresponding to the output results that meet the preset evaluation conditions are used as the basis for resource scheduling, and the heterogeneous resources are then scheduled accordingly.

[0004] In one implementation, each heterogeneous resource is associated with a corresponding baseline operating data, and the baseline operating data is associated with a corresponding load-side resource model and a dynamic response characteristic model. Based on the real-time collected current operating data of each heterogeneous resource, the variable coefficients of the load-side resource model and the variable coefficients of the dynamic response characteristic model of each heterogeneous resource in the virtual power plant are updated. This includes: if the data difference between each current operating data and the corresponding baseline operating data is greater than or equal to a preset difference, then based on the current operating data, the variable coefficients of the load-side resource model and the dynamic response characteristic model associated with the baseline operating data of each heterogeneous resource are updated to obtain the load-side resource model and dynamic response characteristic model of each heterogeneous resource after the variable coefficient update; if the data difference between each current operating data and the corresponding baseline operating data is less than a preset difference, then the variable coefficients of the load-side resource model and the dynamic response characteristic model associated with the baseline operating data of each heterogeneous resource are updated to determine the load-side resource model and dynamic response characteristic model of each heterogeneous resource after the variable coefficient update.

[0005] In another implementation, the load-side resource model includes an electrochemical energy storage model and a distributed photovoltaic power prediction model; the electrochemical energy storage model includes: a total cost model, an operating characteristic model, and a battery life model; the distributed photovoltaic power prediction model characterizes the correlation between photovoltaic power output and meteorological environmental indicators; The total cost model is represented by the following formula:

[0006] in, The total cost of the entire electrochemical energy storage process; This refers to the initial investment cost, including battery purchase, transportation, and installation fees. The operating and maintenance costs include regular replacement and repair expenses, where t represents time; Costs of decommissioning and recycling, including the handling and recycling costs after battery decommissioning; The expression for the running characteristic model is as follows:

[0007]

[0008] in, Indicates the energy efficiency of electrochemical energy storage. and These represent the energy output and input, respectively. Indicates the ramp rate of electrochemical energy storage. and The power at the beginning and end are respectively. Indicates the time of power change; The battery life model expression is as follows:

[0009]

[0010] in, Let the battery degradation function be... It is the initial capacity of the battery. It is the attenuation coefficient. For usage time; To determine the effective lifespan of the battery, a battery capacity threshold is set. When the performance falls below the threshold, it is considered to have reached the end of its lifespan.

[0011] In another implementation, the load-side resource model also includes a heterogeneous load power model; the heterogeneous load power model is characterized by the following formula:

[0012]

[0013]

[0014] The power of the industrial load is , The non-adjustable part This is the adjustable part. For control signals; the power of the commercial load is , Operating hours for business load, It is a comprehensive evaluation index of equipment status; the power of residential loads is For the daily random function of the residents' load, This refers to flexible loads within the residential load.

[0015] In another implementation, the dynamic response characteristic model includes a resource economic response capacity model, a resource response speed model based on incentive intensity, a resource controllability level quantification model, and an operational status perception and correction model; the resource economic response capacity model is characterized by the following formula:

[0016]

[0017]

[0018] Among them, resource economic response capacity , This refers to the maximum adjustable power of the resources, provided that technical conditions permit. The driving coefficient of the excitation intensity on the power response depends on the user's economic sensitivity; The length of time that resources can sustain power reduction or increase. The driving coefficient of excitation intensity on settling time; To the strength of market incentives, and These are the sensitivity coefficients for power response and regulation time response, respectively, and their values ​​differ for different resources; The resource response speed model based on the incentive intensity is characterized by the following formula:

[0019] Among them, the resource response speed is , Basic response speed is the response time of resources under existing conditions; The technology improvement coefficient is the factor that leads users to invest in rapid response technologies due to increased incentive intensity. The resource controllability level quantification model is based on the following formula:

[0020]

[0021]

[0022] in, This represents the device's current maximum adjustable power. The available duration in the current state; This refers to the non-adjustable load ratio constrained by the process or equipment. This represents the current level of market incentives; The price elasticity coefficient reflects the sensitivity of resources to incentives; The adjustment comfort threshold is acceptable to the user; The operational status awareness correction model is characterized by the following formula:

[0023] in, This is a correction factor for load operating status. It is an environmental correction factor.

[0024] In another implementation, the feasible region includes power balance constraints, power flow constraints, voltage stability constraints, and frequency stability constraints. Power balance constraints are expressed by the following formula:

[0025] Power flow constraints are expressed by the following formula:

[0026]

[0027] Voltage stability constraints are expressed by the following formula:

[0028] Frequency stability constraints are expressed by the following formula:

[0029] in, The output power of resources within the virtual power plant. For the load power associated with the machines within the virtual power plant, This refers to the interaction power between the virtual power plant and the external power grid. Let be the active power of line ij; and These are the voltage magnitudes at nodes i and j, respectively; The voltage phase angle difference between nodes; and These are the admittance parameters of the line; This is the frequency offset. and These represent the changes in power generation and load power, respectively. This is the frequency sensitivity coefficient.

[0030] In another implementation, the method further includes: converting the high-dimensional operating space S of the feasible domain into the controllable response capability R of a virtual power plant based on the following formula;

[0031] Furthermore, based on the operating space, determine the upward adjustable power of the virtual power plant that satisfies the constraints. and downward adjustable power .

[0032] In another implementation, based on the operating space, the upward and downward adjustable power of the virtual power plant that meets the constraints is determined, including: Given that the operating space is a convex set, the convex optimization method characterized by the following formula is used for projection to determine the upward and downward adjustable power of the convex set.

[0033]

[0034] Determine the running space For non-convex sets, a convex relaxation method is used to convexize the non-convex constraints. Then, a heuristic method is used to approximate the projection of the convexized non-convex constraints to obtain the upward and downward adjustable power of the non-convex set.

[0035] In another implementation, the state space of the multimodal deep model includes the current electricity price, the available capacity of the virtual power plant, the system load forecast results, and the value at risk; the action space is the combination of output and bid in the energy market and frequency regulation market; and the reward function is a weighted value that comprehensively considers economic benefits and risk penalties.

[0036] According to a second aspect of the present invention, a virtual power plant resource scheduling device is provided, comprising: an update unit, configured to update the variable coefficients of the load-side resource model and the variable coefficients of the dynamic response characteristic model of each heterogeneous resource in the virtual power plant based on real-time collected current operating data of each heterogeneous resource; the load-side resource model includes a representation of the relationship between the resource operating status of each heterogeneous resource and the corresponding physical characteristic indicators and resource environment, and the dynamic response model includes a representation of the dynamic relationship between the economic response indicators of the heterogeneous resources and the user behavior economic indicators and the electricity market environment; The aggregation unit is used to analyze, standardize, and aggregate the variable parameters and relationships of various heterogeneous resources at various time scales in the load-side resource model and dynamic response characteristic model after variable coefficient updates, so as to obtain a multi-time-series aggregation scheduling model. The uncertainty processing unit is used to process multivariate uncertainty indicators based on the bibliometric optimization algorithm, with the minimum revenue indicator as the objective, to obtain the feasible domain of the virtual power plant in each scheduling cycle; the multivariate uncertainty indicators include renewable energy output, load fluctuation and market electricity price; The load forecasting unit is used to predict the load trend of the total load of a virtual power plant represented by multi-source data using a multi-modal deep model, and obtain the load forecast curve and uncertainty range for the current dispatching cycle. The multi-source data includes load data, meteorological data, time feature data, traffic flow data, and event calendar data. The multi-modal deep model is a reinforcement learning model. The evaluation unit is used to generate different inputs to the multi-time series aggregation scheduling model based on the feasible region, load forecast curve and uncertainty interval, and to evaluate the output results of the multi-time series aggregation scheduling model corresponding to different inputs. The scheduling unit is used to use the resource index parameters of each heterogeneous resource corresponding to the output result that meets the preset evaluation conditions as the basis for resource scheduling, and to schedule each heterogeneous resource.

[0037] According to a third aspect of the present invention, a virtual power plant resource scheduling system is provided, the system being configured to perform a virtual power plant resource scheduling method as described in the first aspect and any possible implementation thereof.

[0038] According to a fourth aspect of the present invention, an electronic device is provided, comprising: a processor and a memory for storing processor-executable instructions; wherein the processor is configured to execute the executable instructions to implement a virtual power plant resource scheduling method as described in the first aspect and any possible implementation thereof.

[0039] According to a fifth aspect of the present invention, a computer-readable storage medium is provided, on which instructions are stored, such that when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is able to perform a virtual power plant resource scheduling method as described in the first aspect and any possible implementation thereof.

[0040] According to a sixth aspect of the present invention, a computer program product is provided, the computer program product including computer instructions, which, when executed on an electronic device, cause the electronic device to perform the virtual power plant resource scheduling method of the first aspect and any possible implementation thereof.

[0041] The technical solution provided by this invention brings at least the following beneficial effects: First, based on real-time operational data, the variable coefficients of the constructed load-side resource model and dynamic response characteristic model are updated in real time, so that the updated resource model and characteristic model are more adapted to the current operating environment and operating status, thereby improving the accuracy of the two models. Furthermore, according to multiple time scales, the updated resource model and response characteristic model are analyzed and fused, integrating the physical, environmental, and economic indicators of various heterogeneous resources, so that the multi-time-series aggregated scheduling model simultaneously represents the level of technical and economic controllability. Second, based on the bibliometric optimization algorithm, various uncertainties such as wind and solar power output, load response, and market prices are considered, and optimization is performed on the "worst-case" represented by the minimum return indicator, so that the generated feasible region is a credible and reliable commitment, meaning that even if the situation worsens, the virtual power plant guarantees reasonable scheduling and control within this range. Furthermore, a multimodal deep model using reinforcement learning algorithms is employed to predict the overall load trend of the virtual power plant's total load as reflected by multidimensional source data. The load trend is characterized based on the load prediction curve and the uncertain interval consisting of the load prediction curve and the adjustable range of the load on the load prediction curve. This allows the input load prediction curve to be modified according to the uncertain interval when subsequent adjustments are made based on the overall load trend. In this way, the range of dynamic prediction curves and the feasible region for dynamic adjustment formed by the uncertain interval of the virtual power plant's operating environment can be more rationally used for the scheduling and control of various heterogeneous resources in the virtual power plant.

[0042] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0043] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application, and do not constitute an undue limitation of this application.

[0044] Figure 1 This is a flowchart illustrating a virtual power plant resource scheduling method according to an exemplary embodiment; Figure 2 This is a block diagram illustrating a virtual power plant resource scheduling device according to an exemplary embodiment. Detailed Implementation

[0045] To enable those skilled in the art to better understand the technical solutions of this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0046] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0047] Before providing a detailed introduction to the virtual power plant resource scheduling method provided in the embodiments of this application, let's briefly introduce the application scenarios and implementation environment involved in the embodiments of this application.

[0048] First, a brief introduction to the application scenarios involved in this application will be given.

[0049] Virtual power plants still face a series of severe technical challenges in actual operation and market application. The core problem is that traditional static and isolated modeling and optimization methods are difficult to adapt to the inherent high dynamism and uncertainty of load-side resources, which are specifically manifested in the following aspects: First, there is the mismatch between static resource models and dynamic operating states. Existing virtual power plant technologies typically establish fixed response characteristic models (such as adjustable capacity, response time, and duration) for various load resources (e.g., industrial equipment, electric vehicle charging stations, commercial building air conditioning) based on offline identification or historical data. However, the actual operating states of these resources are constantly changing. For example, factors such as the access status and battery level of electric vehicles, the process stages of industrial production, and the computing load of data centers can all cause significant deviations between their actual regulation capabilities and the static models. This "model inaccuracy" directly leads to a misunderstanding of the virtual power plant's own controllability, making all subsequent optimization and scheduling instructions like "blind men touching an elephant," facing huge deviation risks during actual execution, and even triggering penalties for breach of contract against the power grid.

[0050] Second, there is a lack of robustness in handling multiple uncertainties. The operating environment of virtual power plants is fraught with uncertainty, mainly including: the volatility of distributed renewable energy output (such as photovoltaic and wind power), the randomness of load-side resource response behavior, and the drastic fluctuations in electricity market prices. Existing technologies either employ deterministic optimization (ignoring uncertainty) or overly conservative stochastic planning (leading to poor economic efficiency), making it difficult to achieve an effective balance between economic efficiency and reliability. When the actual operating scenario deviates from the preset conditions, the aggregation dispatch scheme based on traditional methods can easily become infeasible or uneconomical, lacking resilience to withstand risks and seriously affecting the credibility of virtual power plants as reliable market players.

[0051] Third, there is the bottleneck of accuracy in short-term load forecasting. Accurate load forecasting is the foundation for virtual power plants to participate in day-ahead market bidding and formulate internal dispatch plans. Existing forecasting methods mainly rely on historical load curves and conventional meteorological data (such as temperature and humidity), and the models mostly use traditional time series algorithms or single machine learning models. However, the total load of virtual power plants, which consists of a large number of adjustable loads, has more complex variation patterns, especially affected by non-meteorological factors such as traffic flow and large-scale social events. Traditional methods have limitations in data utilization and model structure, and cannot effectively integrate multi-source heterogeneous data, making it difficult to overcome the existing bottleneck in the accuracy of load forecasting, especially impact loads, thus restricting the accuracy of market strategies.

[0052] Fourth, there is a lack of adaptability and risk control capabilities in market participation strategies. When participating in the electricity and ancillary services market, existing virtual power plant bidding strategies largely rely on rule bases, simple price response functions, or static optimization models based on historical data. These strategies lack the ability to learn and dynamically adjust in complex, non-stationary market environments. Furthermore, they often focus on maximizing expected returns while neglecting the potential for significant losses, failing to respond quickly and robustly to drastic changes in market conditions (such as price spikes or rule adjustments), leading to unstable long-term returns and making it difficult to achieve a comprehensive optimization of returns and risks.

[0053] In summary, existing virtual power plant technologies suffer from inherent deficiencies in real-time sensing, uncertainty modeling, multi-source forecasting, and intelligent decision-making, severely limiting their overall operational efficiency, market competitiveness, and reliability. Therefore, there is an urgent need in this field for an intelligent technology solution that can span the entire "sensing-modeling-prediction-decision-making" chain to systematically address these issues and fully unleash the flexible potential of load-side resources.

[0054] To address the aforementioned issues, this application proposes a virtual power plant resource scheduling method. First, it analyzes the operational mechanisms and energy consumption characteristics of heterogeneous resources on a single load side and constructs a load-side resource analytical model library. Simultaneously, it builds a typical response scenario library for load-side resources and establishes a dynamic response characteristic analysis method, constructing a resource interaction response model that considers environmental information. Based on this, for virtual power plant units aggregating multiple resource types, it constructs multi-time-series aggregation modeling and methods for calculating regulation costs and response potential, realizing the projection characterization of the high-dimensional operating space of resources onto the feasible domain of the virtual power plant. Furthermore, it constructs a dynamic parameter correction method to accurately quantify the operational and response characteristics across multiple time series. Finally, based on the reliable quantification results of the virtual power plant unit, it constructs a rolling optimization scheduling strategy of "day-ahead-intraday-real-time," and builds a virtual power plant control command decomposition and optimization model that considers distributed renewable energy and electricity load fluctuation characteristics, thereby providing key technical support for improving the interaction and allocation efficiency of load-side resources.

[0055] For ease of understanding, the virtual power plant resource scheduling method provided in this application will be described in detail below with reference to the accompanying drawings.

[0056] Figure 1 This is a flowchart illustrating a virtual power plant resource scheduling method according to an exemplary embodiment.

[0057] S11, based on the real-time collected current operating data of each heterogeneous resource, update the variable coefficients of the load-side resource model and the variable coefficients of the dynamic response characteristic model of each heterogeneous resource in the virtual power plant.

[0058] The load-side resource model, also known as the load-side resource analytical model library, includes the relationship between the resource operation status of various heterogeneous resources and their corresponding physical characteristic indicators and resource environment.

[0059] The dynamic response model includes the dynamic relationship between economic response indicators representing heterogeneous resources, economic indicators of user behavior, and the electricity market environment.

[0060] Dynamic response model, also known as dynamic response interaction model.

[0061] S12, according to multiple time scales, analyzes, standardizes and aggregates the variable parameters and variable relationships of each heterogeneous resource at each time scale in the load-side resource model and dynamic response characteristic model after variable coefficient update, and obtains a multi-time-series aggregated scheduling model.

[0062] As one implementation method, mean normalization is performed based on variable weights to standardize variable parameters and ensure that variable parameters are consistent in magnitude.

[0063] S13, based on the bibliometric optimization algorithm, takes the minimum revenue index as the objective and processes the multivariate uncertainty index to obtain the feasible domain of the virtual power plant in each scheduling cycle.

[0064] Multiple uncertainty indicators include renewable energy output, load fluctuations, and market electricity prices.

[0065] S14 uses a multimodal deep model to predict the load trend of the total load of the virtual power plant represented by multi-source data, and obtains the load prediction curve and uncertainty range for the current dispatch cycle.

[0066] The multi-source data includes load data, meteorological data, time feature data, traffic flow data, and event calendar data; the multimodal deep model is a reinforcement learning model.

[0067] S15. Based on the feasible region, load forecast curve and uncertainty interval, different inputs to the multi-time series aggregation scheduling model are generated, and the output results of the multi-time series aggregation scheduling model corresponding to different inputs are evaluated.

[0068] S16: The resource index parameters of each heterogeneous resource corresponding to the output results that meet the preset evaluation conditions are used as the basis for resource scheduling to schedule each heterogeneous resource.

[0069] Through the above implementation methods, firstly, based on real-time operational data, the variable coefficients of the constructed load-side resource model and dynamic response characteristic model are updated in real time to make the updated resource model and characteristic model more adaptable to the current operating environment and operating status, thereby improving the accuracy of the two models. Then, according to multiple time scales, the updated resource model and response characteristic model are analyzed and fused, integrating the physical, environmental, and economic indicators of various heterogeneous resources, so that the multi-time-series aggregated scheduling model simultaneously represents the level of technical and economic controllability. Secondly, based on the bibliometric optimization algorithm, various uncertainties such as wind and solar power output, load response, and market prices are considered, and optimization is performed on the "worst-case" scenario represented by the minimum return indicator, so that the generated feasible region is a credible and reliable commitment, meaning that even if the situation worsens, the virtual power plant guarantees reasonable scheduling and control within this range. Furthermore, a multimodal deep model using reinforcement learning algorithms is employed to predict the overall load trend of the virtual power plant's total load as reflected by multidimensional source data. The load trend is characterized based on the load prediction curve and the uncertain interval consisting of the load prediction curve and the adjustable range of the load on the load prediction curve. This allows the input load prediction curve to be modified according to the uncertain interval when subsequent adjustments are made based on the overall load trend. In this way, the range of dynamic prediction curves and the feasible region for dynamic adjustment formed by the uncertain interval of the virtual power plant's operating environment can be more rationally used for the scheduling and control of various heterogeneous resources in the virtual power plant.

[0070] As one implementation method, each heterogeneous resource is associated with a corresponding baseline operating data, and the baseline operating data is associated with a corresponding load-side resource model and a dynamic response characteristic model.

[0071] Based on this, the specific steps for updating the variable coefficients according to the real-time collection of current operating data of various heterogeneous resources include the following steps.

[0072] First, if the difference between each current operating data and the corresponding baseline operating data is greater than or equal to a preset difference, then based on the current operating data, the variable coefficients of the load-side resource model and dynamic response characteristic model associated with the baseline operating data of each heterogeneous resource are updated to obtain the load-side resource model and dynamic response characteristic model of each heterogeneous resource after the variable coefficient update.

[0073] Based on this, the current operating data is associated with the load-side resource model and dynamic response model of each heterogeneous resource after the variable coefficients are updated, and the baseline operating data is updated to the current operating data to ensure the timeliness and accuracy of the baseline operating data for subsequent reference.

[0074] Secondly, if the difference between each current operating data and the corresponding baseline operating data is less than the preset difference, then the load-side resource model and dynamic response characteristic model associated with the baseline operating data of each heterogeneous resource are updated with variable coefficients, and determined as the load-side resource model and dynamic response characteristic model of each heterogeneous resource after the variable coefficient update.

[0075] In one implementation, the load-side resource model includes an electrochemical energy storage model and a distributed photovoltaic power prediction model.

[0076] The above-mentioned electrochemical energy storage model comprehensively considers factors such as the cost, operating characteristics, and service life of electrochemical energy storage devices to construct its mathematical model. The model is used to evaluate the overall economic benefits of electrochemical energy storage over its life cycle and to provide decision support for system design and optimization.

[0077] Specifically, the aforementioned electrochemical energy storage model includes: a total cost model, an operating characteristic model, and a battery life model. Among them, the distributed photovoltaic power prediction model characterizes the correlation between photovoltaic power output and meteorological environmental indicators.

[0078] The total cost model is represented by the following formula.

[0079]

[0080] in, The total cost of the entire electrochemical energy storage process; This refers to the initial investment cost, including battery purchase, transportation, and installation fees. The operating and maintenance costs include regular replacement and repair expenses, where t represents time; Costs of decommissioning and recycling include the handling and recycling fees after battery decommissioning.

[0081] Initial investment costs primarily include the cost of the battery itself and its auxiliary equipment (such as battery management systems and protection circuits). Transportation and installation are also significant components of the cost. During battery use, regular inspections, repairs, and potential component replacements constitute operating and maintenance costs, which are typically closely related to the battery's usage frequency and operating environment. At the end of the battery's lifespan, safe retirement and resource recycling are necessary, which also incurs corresponding economic burdens.

[0082] The efficiency of electrochemical energy storage is a key indicator of its operational characteristics, directly affecting the ratio of actual energy output to input. In practical applications, it is typically necessary to adjust the current and voltage output according to the battery's charge and discharge strategy to meet system requirements. For example, when used in conjunction with load-side resources such as renewable energy, the battery needs to respond quickly to load changes and perform rapid charge and discharge operations, which usually involves power adjustments, thus affecting the battery's energy efficiency and lifespan. Furthermore, the battery's "ramp-up" performance—its ability to rapidly ramp up from low to high power output in a short period—is a crucial performance indicator for electrochemical energy storage systems, especially in grid regulation applications. This requires the battery to provide sufficient response speed and stable energy output to ensure the overall stable operation of the system.

[0083] The operating characteristic model expression for electrochemical energy storage is as follows.

[0084]

[0085]

[0086] in, Indicates the energy efficiency of electrochemical energy storage. and These represent the energy output and input, respectively. Indicates the ramp rate of electrochemical energy storage. and The power at the beginning and end are respectively. This indicates the time of power change.

[0087] Battery degradation is typically caused by a variety of factors, including electrochemical aging, mechanical fatigue, and temperature effects. Degradation models must consider not only the battery's own physicochemical properties but also the influence of external operating conditions. By establishing accurate degradation prediction models, the effective lifespan of the battery can be effectively predicted, and reasonable maintenance and replacement strategies can be formulated accordingly, thereby extending the overall lifespan of the battery system.

[0088] The battery lifetime model expression for electrochemical energy storage is as follows.

[0089]

[0090]

[0091] in, Let the battery degradation function be... It is the initial capacity of the battery. It is the attenuation coefficient. For usage time; To determine the effective lifespan of the battery, a battery capacity threshold is set. When the performance falls below the threshold, it is considered to have reached the end of its lifespan.

[0092] Based on the three electrochemical energy storage models mentioned above, a comprehensive evaluation model for electrochemical energy storage systems can be obtained, which combines cost, operational performance, and lifespan to assess the overall performance of energy storage.

[0093] This model quantifies the specifics of electrochemical energy storage by modeling its cost, performance, and lifespan. It can improve the efficiency of coordination between electrochemical energy storage systems and load-side resources through optimization strategies (such as changing charging and discharging strategies and adjusting maintenance frequency), ultimately achieving optimal economic efficiency and reliability of the energy system.

[0094] Distributed photovoltaic (PV) power prediction models predict the power output of PV systems over a certain period by analyzing historical data, which helps in power dispatch and resource optimization. Distributed PV power prediction models are constructed and adjusted through data collection, preprocessing, feature extraction, model selection and training, and model evaluation and optimization.

[0095] Based on the above prediction models for load-side resource characteristics, electrochemical energy storage, and distributed photovoltaics, and taking into full account the coupling relationship between underlying devices, the analytical constraints of load-side resource operation are analyzed, and a load-side resource model is further constructed.

[0096] In one implementation, the load-side resource model further includes a heterogeneous load power model; the heterogeneous load power model is characterized by the following formula.

[0097]

[0098]

[0099]

[0100] The power of the industrial load is , The non-adjustable part This is the adjustable part. For control signals; the power of the commercial load is , Operating hours for business load, It is a comprehensive evaluation index of equipment status; the power of residential loads is For the daily random function of the residents' load, This refers to flexible loads within the residential load.

[0101] The impact of load-side resources on the external characteristics of a virtual power plant is mainly reflected in its interaction with the power grid or market, particularly in peak shaving and valley filling capabilities, response speed, and economic efficiency. Load-side resource models also include models of the external characteristics of heterogeneous loads.

[0102] The heterogeneous load external characteristic model includes a virtual power plant load-side peak shaving and valley filling capacity model characterized by the following formula.

[0103]

[0104]

[0105] The heterogeneous load external characteristic model also includes a virtual power plant load-side response speed model characterized by the following formula.

[0106]

[0107] The heterogeneous load external characteristic model also includes a virtual power plant load-side economic model characterized by the following formula.

[0108]

[0109] Different feasible domains can be set for the three load types mentioned above.

[0110] Firstly, for industrial loads, industrial equipment such as motors and compressors start and stop frequently, requiring high responsiveness in energy supply. Distributed energy storage can rapidly provide or absorb energy, enhancing system responsiveness in conjunction with photovoltaic systems. Secondly, in industrial production, equipment demands extremely high power stability; any power outage or power quality issue can cause production line shutdowns. Distributed energy storage can serve as a backup power source, providing immediate power when the main grid fails, and combined with distributed photovoltaic systems for continuous power supply, significantly improving system reliability.

[0111] Under this coupling mechanism, the energy balance constraint equation based on the factory load demand is as follows.

[0112]

[0113] in, The total load requirement for normal factory production; The output power of distributed photovoltaic power; and These are the charging and discharging power of the energy storage system, respectively. This refers to the power supplied from the power grid.

[0114] In the coupling mode of industrial load composition, the ramp-up constraint needs to comprehensively consider the ramp-up rate limit of industrial equipment load demand in a short period of time and the operation status of distributed photovoltaic and energy storage at that time. When the grid needs the industrial load to increase its power demand, the industrial equipment increases its power consumption, the distributed photovoltaic decreases its output, and the energy storage system charges; when the grid needs the industrial load to decrease its power demand, the industrial equipment decreases its power consumption, the distributed photovoltaic increases its output, and the energy storage system discharges. The upper and lower limit constraint formulas for the ramp-up rate are as follows.

[0115]

[0116] in, The ramp rate of industrial load under the coupling mechanism; and These are the upper and lower limits of the rate of change of industrial equipment load, respectively. and These are the upper and lower limits of the rate of change of distributed photovoltaic power output, respectively. and These represent the rates of change in charging and discharging of the energy storage system, respectively.

[0117] Accordingly, under the premise of ensuring the normal and economical production operation of the factory, the upper and lower limits of the system's total load demand on the power grid are constrained as follows:

[0118] in, and These are the upper and lower limits of industrial equipment load, respectively, under the premise of meeting economic production requirements; and These represent the maximum charging and discharging power of the energy storage system.

[0119] Secondly, regarding the load on commercial buildings, similar to the industrial load situation described above, the total energy consumption of commercial buildings should be met by photovoltaic power generation, energy storage release, and necessary grid power supply. The ramp-up rate and the upper and lower limits of the overall load depend on the combined effect of the ground floor equipment and photovoltaic and energy storage. Therefore, apart from the output of the ground floor equipment of commercial buildings being limited by the comfort of the building's interior, the rest is similar to the industrial load, and the corresponding energy balance equations and constraint formulas will not be repeated.

[0120] Thirdly, regarding the energy balance of electric vehicle charging stations, the energy supply and demand relationship of charging stations is more complex than that of fixed loads. It is necessary to consider the operating status of distributed photovoltaic and energy storage systems while taking into account their own load demand and the returned power. The energy balance equation is as follows.

[0121]

[0122] in, The total power requirement for electric vehicle charging stations This refers to the power returned to the grid from the charging station.

[0123] Because V2G allows electric vehicles to discharge into the grid rapidly, power stations may need to adjust large amounts of power in a short period, especially during sudden grid demand or price changes. However, frequent charging and discharging of electric vehicle batteries can affect their lifespan and performance. Therefore, although theoretically power stations can adjust power quickly, certain limitations are necessary in practice to prevent excessive battery degradation. Since V2G charging stations simultaneously generate and consume electricity from the grid, the analysis of their load limits is transformed into an analysis of constraints on maximum load and maximum output. The formulas for the maximum load and maximum output constraints are as follows.

[0124]

[0125]

[0126]

[0127] in, and These represent the maximum load and maximum output of the charging station's coupled system, respectively. and These are the maximum charging power and maximum discharge return power of the charging station, respectively.

[0128] In some implementations, the dynamic response characteristic model includes a resource economic response capacity model, an incentive intensity resource response speed model, a resource controllability level quantification model, and an operational status perception correction model.

[0129] The resource-economic response capacity model is based on the following formula.

[0130]

[0131]

[0132]

[0133] Among them, resource economic response capacity , This refers to the maximum adjustable power of the resources, provided that technical conditions permit. The driving coefficient of the excitation intensity on the power response depends on the user's economic sensitivity; The length of time that resources can sustain power reduction or increase. The driving coefficient of excitation intensity on settling time; To the strength of market incentives, and These are the sensitivity coefficients for power response and regulation time response, respectively, and their values ​​differ for different resources.

[0134] The resource response speed model based on incentive intensity is characterized by the following formula.

[0135]

[0136] Among them, the resource response speed is , Basic response speed is the response time of resources under existing conditions; The technology improvement coefficient is the factor that leads users to invest in rapid response technologies due to increased incentive intensity.

[0137] The resource controllability level quantification model is based on the following formula.

[0138]

[0139]

[0140]

[0141] in, This represents the device's current maximum adjustable power. The available duration in the current state; This refers to the non-adjustable load ratio constrained by the process or equipment. This represents the current level of market incentives; The price elasticity coefficient reflects the sensitivity of resources to incentives; The threshold for user-acceptable adjustment comfort.

[0142] The operational status awareness correction model is characterized by the following formula.

[0143]

[0144] in, This is a correction factor for load operating status. It is an environmental correction factor.

[0145] Based on this, real-time operating status is perceived through measurement data and environmental information, primarily considering load operating status measurement data such as real-time power, equipment operating time, load rate, and charging / discharging status, as well as environmental information such as temperature, humidity, and user behavior patterns. Real-time operating status perception is used to dynamically update the load controllability level, and the correction formula for the evaluation result is as follows.

[0146]

[0147] in, This is a correction factor for load operating status. It is an environmental correction factor.

[0148] Based on the following formula, the available capacity interactive response assessment is combined with real-time load status and resource time-sharing characteristics to achieve dynamic assessment. For different types of load, the real-time operating status, correction influencing factors, and interactive evaluation indicators are slightly different when constructing the dynamic response interactive model.

[0149]

[0150] In some implementations, the feasible domain also includes power balance constraints, power flow constraints, voltage stability constraints, and frequency stability constraints.

[0151] High-dimensional network constraints mainly refer to the operational constraints of the power grid in which the virtual power plant is located, including power balance, power flow distribution, and voltage stability. These constraints are usually described by nonlinear physical models and need to be simplified through methods such as equivalence or convex hull transformation. To construct the final virtual power plant projection model, it is necessary to convert the high-dimensional physical constraints into the dynamic operating space of the virtual power plant through power flow calculations and power balance constraint analysis. High-dimensional network constraints can be divided into four key parts: power balance constraints, power flow constraints, voltage stability constraints, and frequency stability constraints of the power grid.

[0152] The power balance constraint is expressed by the following formula.

[0153]

[0154] Power flow constraints are expressed by the following formula.

[0155]

[0156]

[0157] The voltage stability constraint is expressed by the following formula.

[0158]

[0159] The frequency stability constraint is expressed by the following formula.

[0160]

[0161] in, The output power of resources within the virtual power plant. For the load power associated with the machines within the virtual power plant, This refers to the interaction power between the virtual power plant and the external power grid. Let be the active power of line ij; and These are the voltage magnitudes at nodes i and j, respectively; The voltage phase angle difference between nodes; and These are the admittance parameters of the line; This is the frequency offset. and These represent the changes in power generation and load power, respectively. This is the frequency sensitivity coefficient.

[0162] Based on the above analysis, the high-dimensional network constraints of the aforementioned virtual power plant can be simplified to the following form.

[0163] The power balance constraint simplifies to the following formula.

[0164]

[0165] The power flow constraint is simplified to the following formula.

[0166]

[0167]

[0168] The voltage stability constraint is simplified to the following formula.

[0169]

[0170]

[0171] in, As the reference voltage, This is the sensitivity coefficient of voltage to reactive power. Let be the reactive power at the i-th point.

[0172] The frequency constraint simplifies to the following formula.

[0173]

[0174] In one implementation, based on the simplified results of the aforementioned high-dimensional network constraints, and combined with the operational constraints and economic characteristics of heterogeneous resources on the load side, a flexible and efficient mathematical method is used to project the multi-factor operating space onto the controllable response capability of the virtual power plant. The description of the operating space needs to consider key physical quantities in the actual power system, including the decision variables. This includes: the active power P and reactive power Q injected into the grid by the nodes, the current state of energy (SOC) of the energy storage devices, and the load-side regulation power P during demand response. load The on / off state u of the switching equipment, along with other variables, constitutes a high-dimensional decision space. The operating space is the set of feasible solutions to the decision variables while satisfying all constraints; it is generally described by data formulas and includes both constraints and decision variables. Constraints are divided into equality constraints. and inequality constraints The physical constraints of the operating space can be described in the following form.

[0175]

[0176] The constraints of the operating space describe physical constraints such as high-dimensional network constraints and load-side heterogeneous resource operation constraints, but economic efficiency must also be considered. The economic objective is generally cost minimization or revenue maximization. As a market participant, the core objective of a virtual power plant is to optimize economic efficiency, achieving the goals of reducing operating costs and increasing revenue while satisfying physical constraints. Therefore, the final form of the operating space must simultaneously satisfy both high-dimensional constraints and economic objectives. Combining the economic objective with high-dimensional constraints, the operating space is described by the following formula.

[0177]

[0178]

[0179]

[0180] in, Total operating costs, including the cost of purchasing electricity from the market. Operating costs of resources such as energy storage and load response ; For total revenue, including Compensation from participating in demand response and revenue from participating in the electricity and ancillary services markets .

[0181] As one implementation method, the high-dimensional operating space S of the feasible region is transformed into the controllable response capability R of a virtual power plant based on the following formula. The controllable response capability of a virtual power plant is the range of power that it can adjust upward or downward under a given operating space, and its flexibility. The goal is to project the complex high-dimensional operating space into a low-dimensional controllable response capability space. The projection problem can be expressed as the following formula.

[0182]

[0183] Furthermore, based on the operating space, determine the upward adjustable power of the virtual power plant that satisfies the constraints. and downward adjustable power .

[0184] In some implementations, the upward and downward adjustable power of a virtual power plant that meets the constraints is determined based on the operating space, including the following two cases.

[0185] First, given that the operating space is a convex set, the convex optimization method characterized by the following formula is used for projection to determine the upward and downward adjustable power of the convex set.

[0186]

[0187] In the operating space In the middle, find all virtual power plants that satisfy the constraints and can adjust their power upwards. and If the running space Since it is a convex set, it can be projected using convex optimization methods and solved using fast convex optimization algorithms (such as second-order cone programming, SOCP), which can efficiently find the optimal solution.

[0188] Second, determine the operating space. For non-convex sets, a convex relaxation method is used to convexize the non-convex constraints. Then, a heuristic method is used to approximate the projection of the convexized non-convex constraints to obtain the upward and downward adjustable power of the non-convex set.

[0189] Specifically, if the running space Since the set is non-convex, convex relaxation and heuristic methods can be used. Convex relaxation involves making the non-convex constraints convex, such as using Lagrange relaxation or convex hull approximation. Heuristic methods use heuristic search algorithms such as genetic algorithms or particle swarm optimization to find approximate projection results. After projection is completed, the following flexibility index of the virtual power plant is defined using the following formula to characterize its response capability.

[0190] Power adjustment range .

[0191] Response time .

[0192] Benefits of unit power regulation .

[0193] The above methods enable flexible and efficient projection of the operating space into the controllable response capability of the virtual power plant, providing reliable operating boundaries and control capabilities for the virtual power plant while taking into account both economy and real-time performance.

[0194] As one implementation method, the state space of the multimodal deep model includes the current electricity price, the available capacity of the virtual power plant, the system load forecast results, and the value at risk; the action space is the combination of output and bid in the energy market and frequency regulation market; and the reward function is a weighted value that comprehensively considers economic benefits and risk penalties.

[0195] To achieve the above functions, the virtual power plant resource scheduling device includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art will readily recognize that, based on the algorithmic steps of the examples described in conjunction with the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0196] This disclosure also provides an embodiment such as Figure 2 The virtual power plant resource scheduling device shown includes: an update unit 21, an aggregation unit 22, an uncertainty processing unit 23, a load forecasting unit 24, an evaluation unit 25, and a scheduling unit 26.

[0197] The update unit 21 is used to update the variable coefficients of the load-side resource model and the variable coefficients of the dynamic response characteristic model of each heterogeneous resource in the virtual power plant based on the real-time collected current operating data of each heterogeneous resource. The load-side resource model includes the relationship between the resource operating status of each heterogeneous resource and the corresponding physical characteristic indicators and resource environment. The dynamic response model includes the dynamic relationship between the economic response indicators of heterogeneous resources and the user behavior economic indicators and the power market environment.

[0198] Aggregation unit 22 is used to analyze, standardize, and aggregate the variable parameters and relationships of each heterogeneous resource at each time scale in the load-side resource model and dynamic response characteristic model after variable coefficient updates, so as to obtain a multi-time-series aggregated scheduling model.

[0199] Uncertainty processing unit 23 is used to process multivariate uncertainty indicators based on the bibliometric optimization algorithm with the minimum revenue indicator as the objective, so as to obtain the feasible domain of the virtual power plant in each scheduling cycle; the multivariate uncertainty indicators include renewable energy output, load fluctuation and market electricity price.

[0200] The load forecasting unit 24 is used to predict the load trend of the total load of the virtual power plant represented by multi-source data using a multi-modal deep model, and obtain the load forecast curve and uncertainty range of the current dispatching cycle; the multi-source data includes load data, meteorological data, time feature data, traffic flow data, and event calendar data; the multi-modal deep model is a reinforcement learning model.

[0201] Evaluation unit 25 is used to generate different inputs to the multi-time series aggregation scheduling model based on the feasible region, load forecast curve and uncertainty interval, and to evaluate the output results of the multi-time series aggregation scheduling model corresponding to different inputs.

[0202] The scheduling unit 26 is used to use the resource index parameters of each heterogeneous resource corresponding to the output result that meets the preset evaluation conditions as the basis for resource scheduling, and to schedule each heterogeneous resource.

[0203] In one implementation, each heterogeneous resource is associated with a corresponding baseline operating data, and the baseline operating data is associated with a corresponding load-side resource model and a dynamic response characteristic model. The update unit 21 is specifically used to: determine that the data difference between each current operating data and the corresponding baseline operating data is greater than or equal to a preset difference, and then update the variable coefficients of the load-side resource model and dynamic response characteristic model associated with the baseline operating data of each heterogeneous resource based on the current operating data, to obtain the load-side resource model and dynamic response characteristic model of each heterogeneous resource after the variable coefficient update; determine that the data difference between each current operating data and the corresponding baseline operating data is less than a preset difference, and then update the variable coefficients of the load-side resource model and dynamic response characteristic model associated with the baseline operating data of each heterogeneous resource, to determine the load-side resource model and dynamic response characteristic model of each heterogeneous resource after the variable coefficient update.

[0204] Regarding the apparatus in the above embodiments, the specific manner in which each unit module performs its operations has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0205] This application also provides a computer-readable storage medium. When the instructions in the computer-readable storage medium are executed by the processor of a virtual power plant resource scheduling device or electronic device, the virtual power plant resource scheduling device or electronic device is able to perform the virtual power plant resource scheduling method as described in any of the possible implementations above. And it can achieve the same technical effect; to avoid repetition, it will not be described again here.

[0206] This application also provides a computer program product, including a computer program or instructions, which are executed by a processor as a virtual power plant resource scheduling method according to any of the possible implementations described above. Furthermore, it achieves the same technical effects, and to avoid repetition, it will not be described again here.

[0207] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0208] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A virtual power plant resource scheduling method, characterized in that, The method includes: Based on the real-time collection of current operating data of each heterogeneous resource, the variable coefficients of the load-side resource model and the variable coefficients of the dynamic response characteristic model of each heterogeneous resource in the virtual power plant are updated. The load-side resource model includes a representation of the relationship between the resource operating status of each heterogeneous resource and the corresponding physical characteristic indicators and resource environment. The dynamic response model includes a representation of the dynamic relationship between the economic response indicators of heterogeneous resources, user behavior economic indicators and the power market environment. According to multiple time scales, the variable parameters and variable relationships of each heterogeneous resource at each time scale in the load-side resource model and the dynamic response characteristic model after the variable coefficients are updated are analyzed, the variable parameters are standardized and aggregated to obtain a multi-time-series aggregated scheduling model. Based on the bibliometric optimization algorithm, with the minimum revenue index as the objective, the multivariate uncertainty index is processed to obtain the feasible domain of the virtual power plant in each scheduling cycle; the multivariate uncertainty index includes renewable energy output, load fluctuation and market electricity price. A multimodal deep model is used to predict the load trend of the total load of the virtual power plant represented by multi-source data, and the load prediction curve and uncertainty range of the current scheduling cycle are obtained. The multi-source data includes load data, meteorological data, time feature data, traffic flow data, and event calendar data. The multimodal deep model is a reinforcement learning model. Based on the feasible region, the load forecast curve, and the uncertainty interval, different inputs to the multi-time series aggregation scheduling model are generated, and the output results of the multi-time series aggregation scheduling model corresponding to the different inputs are evaluated. The resource index parameters of each heterogeneous resource corresponding to the output results that meet the preset evaluation conditions are used as the basis for resource scheduling, and the heterogeneous resources are scheduled accordingly.

2. The method according to claim 1, characterized in that, Each heterogeneous resource is associated with a corresponding baseline operating data, and the baseline operating data is associated with a corresponding load-side resource model and a dynamic response characteristic model. The step of updating the variable coefficients of the load-side resource model and the variable coefficients of the dynamic response characteristic model of each heterogeneous resource in the virtual power plant based on the real-time collected current operating data of each heterogeneous resource includes: If the data difference between each current operating data and the corresponding baseline operating data is greater than or equal to a preset difference, then based on the current operating data, the variable coefficients of the load-side resource model and dynamic response characteristic model associated with the baseline operating data of each heterogeneous resource are updated to obtain the load-side resource model and dynamic response characteristic model of each heterogeneous resource after the variable coefficient update. If the difference between each current operating data and the corresponding baseline operating data is less than a preset difference, then the load-side resource model and dynamic response characteristic model associated with the baseline operating data of each heterogeneous resource are updated with variable coefficients, and determined as the load-side resource model and dynamic response characteristic model of each heterogeneous resource after the variable coefficient update.

3. The method according to claim 2, characterized in that, The load-side resource model includes an electrochemical energy storage model and a distributed photovoltaic power prediction model; the electrochemical energy storage model includes a total cost model, an operating characteristic model, and a battery life model; the distributed photovoltaic power prediction model characterizes the correlation between photovoltaic power output and meteorological environmental indicators; The total cost model is represented by the following formula: in, The total cost of the entire electrochemical energy storage process; This refers to the initial investment cost, including battery purchase, transportation, and installation fees. The operating and maintenance costs include regular replacement and repair expenses, where t represents time; Costs of decommissioning and recycling, including the handling and recycling costs after battery decommissioning; The expression for the operational characteristic model is as follows: in, Indicates the energy efficiency of electrochemical energy storage. and These represent the energy output and input, respectively. Indicates the ramp rate of electrochemical energy storage. and Power at the start and end, respectively. Indicates the time of power change; The battery life model expression is as follows: in, Let the battery degradation function be... It is the initial capacity of the battery. It is the attenuation coefficient. For usage time; To determine the effective lifespan of the battery, a battery capacity threshold is set. When the performance falls below the threshold, it is considered to have reached the end of its lifespan.

4. The method according to claim 3, characterized in that, The load-side resource model also includes a heterogeneous load power model; the heterogeneous load power model is characterized by the following formula: The power of the industrial load is , The non-adjustable part This is the adjustable part. For control signals; the power of the commercial load is , Operating hours for business load, It is a comprehensive evaluation index of equipment status; the power of residential loads is For the daily random function of the residents' load, This refers to flexible loads within the residential load.

5. The method according to claim 4, characterized in that, The dynamic response characteristic model includes a resource economic response capacity model, an incentive intensity resource response speed model, a resource controllability level quantification model, and an operational status perception and correction model; the resource economic response capacity model is characterized by the following formula: Among them, resource economic response capacity , This refers to the maximum adjustable power of the resources, provided that technical conditions permit. The driving coefficient of the excitation intensity on the power response depends on the user's economic sensitivity; The length of time that resources can sustain power reduction or increase. The driving coefficient of excitation intensity on settling time; To the strength of market incentives, and These are the sensitivity coefficients for power response and regulation time response, respectively, and their values ​​differ for different resources; The resource response speed model for the incentive intensity is characterized by the following formula: Among them, the resource response speed is , Basic response speed is the response time of resources under existing conditions; The technology improvement coefficient is the factor that leads users to invest in rapid response technologies due to increased incentive intensity. The resource controllability level quantification model is based on the following formula: in, This represents the device's current maximum adjustable power. The available duration in the current state; This refers to the non-adjustable load ratio constrained by the process or equipment. This represents the current level of market incentives; The price elasticity coefficient reflects the sensitivity of resources to incentives; The adjustment comfort threshold is acceptable to the user; The operational status awareness correction model is characterized by the following formula: in, This is a correction factor for load operating status. It is an environmental correction factor.

6. The method according to claim 5, characterized in that, The feasible region includes power balance constraints, power flow constraints, voltage stability constraints, and frequency stability constraints. The power balance constraint is expressed by the following formula: The power flow constraint is expressed by the following formula: The voltage stability constraint is expressed by the following formula: The frequency stability constraint is expressed by the following formula: in, The output power of resources within the virtual power plant. For the load power associated with the machines within the virtual power plant, This refers to the interaction power between the virtual power plant and the external power grid. Let be the active power of line ij; and These are the voltage magnitudes at nodes i and j, respectively; The voltage phase angle difference between nodes; and These are the admittance parameters of the line; This is the frequency offset. and These represent the changes in power generation and load power, respectively. This is the frequency sensitivity coefficient.

7. The method according to claim 6, characterized in that, The method further includes: Based on the following formula, the high-dimensional operating space S of the feasible domain is converted into the controllable response capability R of the virtual power plant; Furthermore, based on the aforementioned operating space, the upward adjustable power of the virtual power plant that satisfies the constraints is determined. and downward adjustable power .

8. The method according to claim 7, characterized in that, Based on the aforementioned operating space, the upward and downward adjustable power of the virtual power plant that satisfy the constraints are determined, including: If the operating space is determined to be a convex set, a convex optimization method characterized by the following formula is used for projection to determine the upward adjustable power and downward adjustable power of the convex set; Determine the running space The non-convex set is made convex by using a convex relaxation method. Then, the convex non-convex constraint is approximated by the heuristic method to obtain the upward adjustable power and downward adjustable power of the non-convex set.

9. The method according to any one of claims 1 to 8, characterized in that, The state space of the multimodal deep model includes the current electricity price, available capacity of the virtual power plant, system load forecast results, and risk value; the action space is the combination of output and bid in the energy market and frequency regulation market; and the reward function is a weighted value that comprehensively considers economic benefits and risk penalties.

10. A virtual power plant resource scheduling device, characterized in that, The device includes: The update unit is used to update the variable coefficients of the load-side resource model and the variable coefficients of the dynamic response characteristic model of each heterogeneous resource in the virtual power plant based on the real-time collected current operating data of each heterogeneous resource. The load-side resource model includes a representation of the relationship between the resource operating status of each heterogeneous resource and the corresponding physical characteristic indicators and resource environment. The dynamic response model includes a representation of the dynamic relationship between the economic response indicators of heterogeneous resources and the user behavior economic indicators and the power market environment. The aggregation unit is used to analyze, standardize, and aggregate the variable parameters and relationships of each heterogeneous resource at each time scale in the load-side resource model and the dynamic response characteristic model after the variable coefficients are updated, so as to obtain a multi-time-series aggregated scheduling model. Based on the bibliometric optimization algorithm, with the minimum revenue index as the objective, the multivariate uncertainty index is processed to obtain the feasible domain of the virtual power plant in each scheduling cycle; the multivariate uncertainty index includes renewable energy output, load fluctuation and market electricity price. The load forecasting unit is used to predict the load trend of the total load of the virtual power plant represented by multi-source data using a multi-modal deep model, and to obtain the load forecast curve and uncertainty range for the current dispatching cycle; the multi-source data includes load data, meteorological data, time feature data, traffic flow data, and event calendar data; the multi-modal deep model is a reinforcement learning model; The evaluation unit is used to generate different inputs to the multi-time series aggregation scheduling model based on the feasible region, the load forecast curve and the uncertainty interval, and to evaluate the output results of the multi-time series aggregation scheduling model corresponding to the different inputs. The scheduling unit is used to use the resource index parameters of each heterogeneous resource corresponding to the output result that meets the preset evaluation conditions as the basis for resource scheduling, and to schedule each heterogeneous resource.

Images