A virtual power plant dynamic resource optimization scheduling method and system

Abstract

The application relates to the technical field of electric power, and discloses a virtual power plant dynamic resource optimization scheduling method and system. The application solves the identification deviation problem caused by parameter coupling, and improves the robustness of a scheduling strategy to parameter uncertainty. An equivalent parameter combination method reduces the model dimension, so that multi-parameter joint estimation has computational feasibility. A multi-time scale optimization framework realizes dynamic cooperation of a parameter model and a scheduling strategy, and avoids scheduling deviation caused by parameter update lag in a traditional method. A coverage rate calibration mechanism ensures the matching degree of an uncertainty set and an actual error distribution, and provides a reliable data basis for robust optimization. The application realizes accurate modeling and identifiable optimization of key parameters of a virtual power plant, and solves the model misalignment problem caused by parameter redundancy or unidentifiability in a traditional method. Through generation of a candidate identifiable parameter set, the feasibility of subsequent experimental design and parameter estimation processes is ensured.

Description

Technical Field

[0001] This invention relates to the field of power technology, specifically to a method and system for dynamic resource optimization scheduling of virtual power plants. Background Technology

[0002] With the large-scale integration of distributed power sources, energy storage devices, and adjustable loads, virtual power plants (VPS) are gradually becoming an important form of power system flexibility as an aggregated control platform. During the operation of VPS, the degradation characteristics of energy storage devices, the adjustable characteristics of user-side loads, the derating characteristics of inverters under environmental conditions, and the elastic response of users to electricity prices all directly affect dispatch effectiveness. Therefore, it is necessary to systematically model and jointly optimize these parameters to ensure the overall operational performance and economy of the VPS.

[0003] In existing technologies, virtual power plant dynamic dispatching methods mostly rely on empirical parameters or simplified models for optimization, lacking systematic identification and consistent joint modeling of energy storage degradation parameters, load utility parameters, inverter derating parameters, and price elasticity parameters. Because these parameters often exhibit unidentifiable characteristics, strong correlations, or equivalence class issues, the dispatching results deviate from actual operating conditions, leading to problems such as unexecutable dispatching plans, increased deviation assessment costs, or excessive equipment degradation. Furthermore, existing methods insufficiently handle parameter uncertainties during the dispatching process, lacking adaptive mechanisms based on uncertain sets and robust optimization, and cannot effectively cope with changes in the operating environment and data quality fluctuations, thus limiting the widespread application of virtual power plants in complex power systems.

[0004] Therefore, we propose a dynamic resource optimization scheduling method for virtual power plants to address the aforementioned problems. Summary of the Invention

[0005] The purpose of this invention is to provide a dynamic resource optimization scheduling method for virtual power plants, which solves the problem that the dynamic scheduling methods for virtual power plants proposed in the background art mostly rely on empirical parameters or simplified models for optimization, and lack the problem of joint modeling with consistent system identification and constraints for energy storage degradation parameters, load utility parameters, inverter derating parameters and price elasticity parameters.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for dynamic resource optimization scheduling of a virtual power plant, the specific steps of which are as follows:

[0007] S1: Perform structured modeling of energy storage degradation parameters, adjustable load utility parameters, inverter thermal derating parameters, and price elasticity parameters; determine the set of identifiable parameters based on structured identifiability analysis, and reparameterize parameters that are unidentifiable or have equivalence classes to obtain a set of candidate identifiable parameters;

[0008] S2: Under the constraints of power grid security, user constraints and market constraints, excitation signals are generated for charging and discharging behavior, load shifting behavior and reactive power settings;

[0009] S3: Construct the optimal experimental design model and solve the stimulus scheme using Fisher's information content or mutual information content criteria;

[0010] S4: Establish a joint estimation model that includes energy storage parameters, load parameters, inverter parameters, and price parameters; introduce grid constraints, user constraints, and market constraints as conditions during the parameter estimation process; calculate the parameter estimates and corresponding confidence intervals using Bayesian inference or profile likelihood methods;

[0011] S5: Combine parameters that are difficult to identify individually into equivalent parameters; generate parameter uncertainty sets based on the estimation results, wherein the uncertainty sets are in the form of ellipsoidal sets, box constraints or polyhedra, and perform coverage calibration; establish upper and lower bound models for the parameters;

[0012] S6: In the multi-timescale scheduling framework of day-ahead, intraday, and real-time, the uncertainty set and degradation model obtained in step 5 are introduced into the scheduling constraints and objectives; an optimization model containing operational objectives and information gain terms is constructed, and the scheduling strategy is obtained by solving the bibliometric optimization method or the chance constraint method.

[0013] S7: After scheduling execution, calibrate the chance constraint coverage; compare and evaluate the parameter identification results with the scheduling scheme; update the experimental design scheme and reparameterized model; generate and store records of the scheduling process and parameter updates.

[0014] Preferably, step S1 is performed in the following manner:

[0015] S1.1: Establish a mathematical relationship model between energy storage degradation parameters, adjustable load utility parameters, inverter thermal derating parameters, and price elasticity parameters and equipment operating status, external input and output variables. Based on the mathematical relationship model, use the structured identifiability analysis method to determine the set of identifiable parameters.

[0016] S1.2: Reparameterize the detected unidentifiable parameters or parameters belonging to the equivalence class to obtain identifiable equivalent parameters, thereby generating a candidate set of identifiable parameters as input for subsequent experimental design and parameter estimation.

[0017] Preferably, step S2 is performed in the following manner:

[0018] S2.1: Under the constraints of power grid security, user constraints and market constraints, determine the feasible range of energy storage charging and discharging, load shifting and reactive power setting, generate excitation signals within the feasible range, and use the excitation signals as experimental inputs;

[0019] S2.2: Construct an optimal experimental design model based on the excitation signal, introduce Fisher's information criterion or mutual information criterion into the model, solve the model, and obtain the optimal excitation scheme that satisfies the constraints.

[0020] Preferably, step S3 is implemented in the following manner:

[0021] S3.1: During the execution of the excitation signal, multi-source operating data is collected, time synchronization processing and event labeling processing are performed on the collected data, and data with missing information is filled in.

[0022] S3.2: Perform anomaly detection and credibility assessment on the collected data, remove data that does not meet the requirements, and generate a processed experimental dataset.

[0023] Preferably, step S4 is performed in the following manner:

[0024] S4.1: Establish a joint estimation model that includes energy storage parameters, load parameters, inverter parameters and price parameters. In the model, grid constraints, user constraints and market constraints are introduced as conditions to form a parameter estimation framework that meets the operating conditions.

[0025] S4.2: Based on the joint estimation model, the parameters are estimated using Bayesian inference or profile likelihood methods, and the estimated values ​​of the parameters and their corresponding confidence intervals are output.

[0026] Preferably, step S5 is implemented in the following manner:

[0027] S5.1: Combine parameters that are difficult to identify individually, combine the parameters into equivalent parameters, and perform reparameterization processing on the parameter set;

[0028] S5.2: Generate a parameter uncertainty set based on the result of the reparameterization process. The uncertainty set is represented in the form of an ellipsoid set, box constraint, or polyhedron. The coverage of the uncertainty set is calibrated, and an upper and lower bound model of the parameters is established.

[0029] Preferably, step S6 is implemented in the following manner:

[0030] S6.1: In the scheduling framework of day-ahead, intraday and real-time multi-timescale, the uncertainty set and degradation model obtained in step S5 are introduced into the scheduling objectives and scheduling constraints to construct an optimization model that includes operational objective terms and information gain terms.

[0031] S6.2: Based on the optimization model, the solution is obtained by using the bibliometric optimization method or the chance constraint method, and the scheduling strategy that satisfies the constraint conditions is output.

[0032] Preferably, step S7 is implemented in the following manner:

[0033] S7.1: After the scheduling is completed, the opportunity constraint coverage is calibrated, and the parameter identification results are compared and evaluated with the scheduling scheme execution.

[0034] S7.2: Update the experimental design scheme and reparameterized model based on the comparative evaluation results, and generate and store scheduling process records and parameter update records.

[0035] This application also includes a virtual power plant dynamic resource optimization scheduling system, comprising a parameter modeling module, an incentive design module, a data acquisition module, a parameter estimation module, an uncertainty set module, a rolling scheduling module, and a verification and update module;

[0036] The parameter modeling module is used to perform structured modeling of energy storage degradation parameters, adjustable load utility parameters, inverter thermal derating parameters and price elasticity parameters, and to determine the set of identifiable parameters based on structured identifiability analysis. Parameters that are not identifiable or belong to equivalence classes are reparameterized to generate a set of candidate identifiable parameters.

[0037] The incentive design module is used to determine the feasible range of charging and discharging behavior, load shifting behavior and reactive power setting under the constraints of power grid security, user constraints and market constraints, construct the optimal experimental design model, solve the model based on Fisher's information criterion or mutual information criterion, and output the incentive scheme.

[0038] The data acquisition module is used to collect multi-source runtime data during the stimulus execution process, perform time synchronization, event labeling and missing data completion on the data, and perform anomaly detection and credibility assessment on the data, remove data that does not meet the requirements, and generate an experimental dataset.

[0039] The parameter estimation module is used to establish a joint estimation model that includes energy storage parameters, load parameters, inverter parameters and price parameters. The model introduces grid constraints, user constraints and market constraints as conditions, and calculates the parameter estimates and their corresponding confidence intervals based on Bayesian inference method or profile likelihood method.

[0040] The uncertainty set module is used to perform reparameterization processing on parameters that are difficult to identify individually, combine the parameters into equivalent parameters, and generate a parameter uncertainty set based on the estimation results. The parameter uncertainty set is represented in the form of an ellipsoid set, box constraint, or polyhedron, and an upper bound model and a lower bound model for the parameters are established.

[0041] The rolling scheduling module is used to introduce the uncertainty set and degradation model into the scheduling objectives and constraints in the multi-timescale scheduling framework of day-ahead, intraday and real-time, to construct an optimization model containing operational objective items and information gain items, and to solve the optimization model by the split-bar optimization method or the chance constraint method to output the scheduling strategy.

[0042] The verification and update module is used to calibrate the chance constraint coverage after the scheduling is completed, compare and evaluate the parameter identification results with the scheduling scheme, update the experimental design scheme and reparameterized model based on the comparison and evaluation, and generate and store scheduling process records and parameter update records.

[0043] The beneficial effects of this invention are:

[0044] 1. This application addresses the identification bias problem caused by parameter coupling, improving the robustness of scheduling strategies to parameter uncertainties. The equivalent parameter combination method reduces model dimensionality, making multi-parameter joint estimation computationally feasible. The multi-timescale optimization framework enables dynamic coordination between the parameter model and the scheduling strategy, avoiding scheduling bias caused by parameter update lag in traditional methods. The coverage calibration mechanism ensures the matching degree between the uncertainty set and the actual error distribution, providing a reliable data foundation for robust optimization.

[0045] 2. This application achieves accurate modeling and identifiability optimization of key parameters in virtual power plants, solving the model inaccuracy problem caused by parameter redundancy or unidentifiable parameters in traditional methods. By generating a set of candidate identifiable parameters, the feasibility of subsequent experimental design and parameter estimation processes is ensured, providing a fundamental guarantee for the accuracy of dynamic resource scheduling in virtual power plants. Attached Figure Description

[0046] Figure 1 This is a diagram illustrating the steps of the method of the present invention.

[0047] Figure 2 This is a system flowchart of the present invention. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Example 1: Please refer to Figure 1 A method for dynamic resource optimization scheduling of virtual power plants, the specific steps of which are as follows:

[0050] S1: Perform structured modeling of energy storage degradation parameters, adjustable load utility parameters, inverter thermal derating parameters, and price elasticity parameters; determine the set of identifiable parameters based on structured identifiability analysis, and reparameterize parameters that are unidentifiable or have equivalence classes to obtain a set of candidate identifiable parameters;

[0051] S2: Under the constraints of power grid security, user constraints and market constraints, excitation signals are generated for charging and discharging behavior, load shifting behavior and reactive power settings;

[0052] S3: Construct the optimal experimental design model and solve the stimulus scheme using Fisher's information content or mutual information content criteria;

[0053] S4: Establish a joint estimation model that includes energy storage parameters, load parameters, inverter parameters, and price parameters; introduce grid constraints, user constraints, and market constraints as conditions during the parameter estimation process; calculate the parameter estimates and corresponding confidence intervals using Bayesian inference or profile likelihood methods;

[0054] S5: Combine parameters that are difficult to identify individually into equivalent parameters; generate parameter uncertainty sets based on the estimation results, and take the form of ellipsoidal sets, box constraints or polyhedra, and perform coverage calibration; establish upper and lower bound models for parameters.

[0055] S6: In the multi-timescale scheduling framework of day-ahead, intraday, and real-time, the uncertainty set and degradation model obtained in step 5 are introduced into the scheduling constraints and objectives; an optimization model containing operational objectives and information gain terms is constructed, and the scheduling strategy is obtained by solving the bibliometric optimization method or the chance constraint method.

[0056] S7: After scheduling execution, calibrate the chance constraint coverage; compare and evaluate the parameter identification results with the scheduling scheme; update the experimental design scheme and reparameterized model; generate and store records of the scheduling process and parameter updates.

[0057] In this embodiment: In existing technologies, virtual power plant dynamic dispatching methods often rely on empirical parameters or simplified models for optimization, lacking systematic identification and consistent joint modeling of energy storage degradation parameters, load utility parameters, inverter derating parameters, and price elasticity parameters. Due to the unidentifiable nature, strong correlation, or equivalence class issues of these parameters, the dispatching results deviate from actual operating conditions, leading to problems such as unexecutable dispatching plans or excessive equipment degradation. Furthermore, existing methods are insufficient in handling parameter uncertainties and cannot effectively cope with changes in the operating environment and fluctuations in data quality. For example, when a virtual power plant operates in a high-temperature environment during summer, the inverter's output is limited due to inaccurate identification of thermal derating parameters, but the existing dispatching model fails to dynamically correct the parameters, resulting in a power supply gap.

[0058] To address the aforementioned issues, the inventors discovered that a key flaw in existing technologies lies in the separation between parameter identification and scheduling optimization. First, energy storage degradation parameters and load utility parameters are coupled, and identifying them separately leads to equivalence class problems. Second, parameter uncertainty is not incorporated into the optimization objective during scheduling, resulting in insufficient policy robustness. Based on this, the inventors propose combining structured parameter modeling with experimental design, actively stimulating the system's dynamic characteristics through excitation signals, and establishing a joint estimation model to resolve the parameter coupling problem. Simultaneously, a set of parameter uncertainties is introduced into multi-timescale scheduling to achieve adaptive adjustment of the scheduling strategy.

[0059] Therefore, this application proposes the following technical solutions: Structured modeling of energy storage degradation parameters, adjustable load utility parameters, inverter thermal derating parameters, and price elasticity parameters; determination of parameter sets based on identifiability analysis and subsequent parameterization; generation of excitation signals under grid, user, and market constraints; construction of an optimal experimental design model; collection and processing of multi-source data to establish a joint estimation model; calculation of parameter estimates using Bayesian inference or profile likelihood methods; combination of unidentifiable parameters into equivalent parameters to generate an uncertainty set; introduction of the uncertainty set into a multi-timescale scheduling framework to construct an optimization model; solution of the scheduling strategy using distributed bar optimization or chance constraint methods; and calibration of coverage and model update after execution.

[0060] Structured modeling refers to establishing a mathematical relationship model between parameters and equipment operating states, external inputs, and output variables. This can be implemented using state equations or transfer functions, and is used to analyze the correlation between parameters. Fisher's information criterion optimizes excitation signals by quantifying the accuracy of parameter estimation. This can be implemented using the information matrix determinant maximization method, and is used to improve parameter identification efficiency. Bayesian inference methods update the posterior distribution of parameters based on prior distributions and observed data. This can be implemented using Markov chain Monte Carlo sampling, and is used to handle parameter uncertainty. Equivalent parameter combination refers to merging strongly correlated parameters into a single representative parameter. This can be implemented using principal component analysis, and is used to solve unidentifiable problems. Wasserstein distance fuzzy set optimization seeks the optimal solution under worst-case conditions within an uncertain set. This can be implemented using Wasserstein distance to construct fuzzy sets, and is used to improve the robustness of scheduling strategies.

[0061] First, mathematical models of energy storage, load, inverter, and price parameters are established, and the structural relationships between parameters are analyzed. Identifiability analysis is used to screen independently identifiable parameters, and parameters with equivalence classes are merged and recombined. Then, a composite signal incorporating charging / discharging excitation, load transfer excitation, and reactive power regulation excitation is designed to maximize the amount of parameter identification information while satisfying grid security constraints. After excitation is executed, multi-source data is collected synchronously, and anomalous data is cleaned before being input into the joint estimation model. This model uses grid operation constraints as boundary conditions and employs probabilistic statistical methods to calculate parameter estimates and confidence intervals. For parameter combinations with overlapping confidence intervals, equivalent parameters are established, and an ellipsoidal uncertainty set is constructed. A parameter degradation model is considered during day-ahead scheduling, and an uncertainty set is introduced into real-time scheduling to construct a bi-objective optimization model, simultaneously optimizing operating costs and information gain. After scheduling execution, the impact of parameter estimation errors on coverage is evaluated, and the experimental design scheme and equivalent parameter combination rules are dynamically updated.

[0062] Existing methods do not consider the impact of grid operation constraints on parameter estimation during the parameter identification stage, leading to a mismatch between offline identified parameters and online operating conditions. This solution embeds grid, user, and market constraints into the parameter estimation process, ensuring that the identification results conform to the actual operating boundaries. Existing technologies employ single-timescale scheduling, which struggles to handle time-varying parameter characteristics; this solution achieves coordinated updates of the parameter model and scheduling strategy through a multi-timescale framework. Existing methods often use fixed boundaries to handle parameter uncertainty; this solution dynamically adjusts the uncertainty set range through coverage calibration, improving adaptability to data quality fluctuations.

[0063] This application addresses the identification bias problem caused by parameter coupling, improving the robustness of scheduling strategies to parameter uncertainties. The equivalent parameter combination method reduces model dimensionality, making multi-parameter joint estimation computationally feasible. The multi-timescale optimization framework enables dynamic coordination between the parameter model and the scheduling strategy, avoiding scheduling biases caused by parameter update lags in traditional methods. The coverage calibration mechanism ensures the matching degree between the uncertainty set and the actual error distribution, providing a reliable data foundation for robust optimization.

[0064] Example 2: Please refer to Figure 1 The specific method for step S1 is as follows:

[0065] S1.1: Establish a mathematical relationship model between energy storage degradation parameters, adjustable load utility parameters, inverter thermal derating parameters, and price elasticity parameters and equipment operating status, external input and output variables. Based on the mathematical relationship model, use the structured identifiability analysis method to determine the set of identifiable parameters.

[0066] S1.2: Reparameterize the detected unidentifiable parameters or parameters belonging to the equivalence class to obtain identifiable equivalent parameters, thereby generating a candidate set of identifiable parameters as input for subsequent experimental design and parameter estimation.

[0067] In this embodiment: This application further proposes a dynamic resource optimization scheduling method for virtual power plants, the steps of which include: performing structured modeling of energy storage degradation parameters, adjustable load utility parameters, inverter thermal derating parameters, and price elasticity parameters; establishing a mathematical relationship model between parameters and equipment operating status, external inputs, and output variables; based on the structured model, using a structured identifiability analysis method to determine the set of identifiable parameters; performing reparameterization processing on detected unidentifiable parameters or parameters belonging to equivalence classes to obtain identifiable equivalent parameters; thereby generating a candidate set of identifiable parameters as input for subsequent experimental design and parameter estimation.

[0068] A mathematical relationship model refers to the formal expression of the causal relationship between parameters and equipment operating variables through mathematical equations or statistical models. Specifically, it can be implemented using state-space equations, differential-algebraic equations, or stochastic process models to establish an explicit association between parameters and the dynamic behavior of the system.

[0069] Structured identifiability analysis refers to evaluating the identifiability of parameters through parameter sensitivity analysis or information matrix rank test methods. Specifically, it can be implemented using symbolic computation or numerical differentiation methods to screen out a subset of parameters that have unique solutions under given observation data.

[0070] Unidentifiable parameters are those that cannot be uniquely determined from existing observation data. Specifically, they are characterized by linear dependence or functional relationships between parameters, leading to multiple solutions in parameter estimation results.

[0071] Equivalence classes refer to different combinations of parameters that have the same input-output response characteristics. Specifically, the contribution of the parameters in the model cannot be distinguished, resulting in equivalent parameter estimation results.

[0072] Reparameterization refers to transforming unidentifiable parameters into identifiable forms through parameter transformation or dimensionality reduction methods. Specifically, it can be achieved using principal component analysis, parameter bundling, or normalization methods to eliminate parameter redundancy and improve estimation efficiency.

[0073] First, a mathematical model is formed by establishing a correlation model between energy storage degradation parameters and charge / discharge cycle count, a mapping relationship between adjustable load utility parameters and user response curves, a coupling equation between inverter thermal derating parameters and ambient temperature, and a dynamic response function between price elasticity parameters and demand changes. Then, the Jacobian matrix rank test method is used to determine the identifiability of the parameters. When a rank deficiency is detected in the parameter sensitivity matrix, the indiscernibility is eliminated by introducing parameter normalization constraints or constructing equivalent parameter combinations. For example, the energy storage capacity degradation coefficient and cycle efficiency parameter are bundled into an equivalent attenuation factor, thereby generating a uniquely identifiable set of candidate parameters.

[0074] Traditional methods typically employ empirical parameters or independent modeling approaches, neglecting the coupling relationships and identifiability issues between parameters. This proposed solution, however, combines structured modeling with identifiability analysis to effectively identify and eliminate parameter redundancy, avoiding model distortion caused by unidentifiable parameters. Common parameter estimation biases in existing technologies are systematically addressed in this solution, providing a reliable foundation for subsequent excitation signal generation and parameter estimation.

[0075] This application achieves accurate modeling and identifiability optimization of key parameters in virtual power plants, solving the model inaccuracy problem caused by parameter redundancy or unidentifiable parameters in traditional methods. By generating a set of candidate identifiable parameters, the feasibility of subsequent experimental design and parameter estimation processes is ensured, providing a fundamental guarantee for the accuracy of dynamic resource scheduling in virtual power plants.

[0076] Example 3: Please refer to Figure 1 The specific method for step S2 is as follows:

[0077] S2.1: Under the constraints of power grid security, user constraints and market constraints, determine the feasible range of energy storage charging and discharging, load shifting and reactive power setting, generate excitation signals within the feasible range, and use the excitation signals as experimental inputs;

[0078] S2.2: Construct an optimal experimental design model based on the excitation signal, introduce Fisher's information criterion or mutual information criterion into the model, solve the model, and obtain the optimal excitation scheme that satisfies the constraints.

[0079] In this embodiment: This application further proposes to determine the feasible range of energy storage charging and discharging, load shifting and reactive power setting under the conditions of grid security constraints, user constraints and market constraints, and generate excitation signals as experimental inputs within the feasible range; construct an optimal experimental design model based on the excitation signals, introduce Fisher information criterion or mutual information criterion into the model, and solve the model to obtain the optimal excitation scheme that satisfies the constraints.

[0080] Grid security constraints refer to voltage, frequency, and line capacity limitations that ensure the stable operation of the power system. These can be described using power flow equations and dynamic stability models to define the physical boundaries of excitation signals. User constraints refer to load-side requirements for power reliability and comfort, such as start-up and shutdown time limits for adjustable loads or power adjustment thresholds, which can be quantified through user agreements or equipment operation specifications. Market constraints refer to the requirements of power trading rules on the range of power adjustments and response speed. For example, the difference in clearing prices between the day-ahead market and the real-time market restricts charging and discharging behavior, which can be modeled through market operation rules. Excitation signals are input signals generated by adjusting the charging and discharging power of energy storage, load transfer, or reactive power setpoints. These can take the form of pulse sequences, step changes, or random disturbances to stimulate the system's dynamic response and enhance parameter identifiability. Fisher's information criterion optimizes experimental design by maximizing the determinant of the Fisher information matrix of the parameter estimates. This can be achieved through matrix inversion and eigenvalue decomposition, and is used to improve the accuracy of parameter estimation. Mutual information criterion optimizes the stimulus scheme by maximizing the mutual information between the input signal and the output response. This can be achieved through information entropy calculation and gradient optimization methods, and is used to enhance the robustness of parameter identification.

[0081] First, based on power grid operation regulations, user electricity agreements, and market trading rules, upper and lower limits for energy storage charging and discharging power, time limits for load shifting, and voltage deviation ranges for reactive power regulation are established, forming a multi-dimensional constraint space. Within this constraint space, a set of candidate excitation signals is generated, encompassing different combinations of charging and discharging rates, load shifting magnitudes, and reactive power settings. Subsequently, an optimal experimental design model is constructed with Fisher information or mutual information as the objective function. Numerical optimization algorithms are used to select the excitation signal sequence that maximizes parameter identification effectiveness. For example, in the energy storage parameter identification scenario, alternating high-power charging and discharging pulses with low-power steady-state operation can effectively stimulate the dynamic response of battery degradation characteristics; in load elasticity parameter estimation, a tiered electricity price excitation signal can be used to observe differences in load shifting behavior across different price ranges.

[0082] Traditional methods typically employ fixed excitation patterns or signal design based on empirical rules, such as periodic charging and discharging or random perturbations, lacking system optimization for parameter identification efficiency. This approach, however, introduces an information content criterion to construct an experimental design model, combining the statistical characteristics of parameter estimation with system constraints. This allows for the generation of optimal excitation signals while ensuring operational safety, avoiding resource waste or equipment damage caused by ineffective experiments. For example, existing technologies using random excitation may lead to increased parameter estimation errors due to mismatches between the signal spectrum and system dynamics. This approach, however, uses the information content criterion to select excitation signals that match the parameter-sensitive frequency bands, significantly improving data validity.

[0083] This application can generate excitation signal sequences that match parameter identification requirements, maximizing the information value of experimental data while satisfying multiple constraints. This method solves the problems of high blindness in traditional excitation design and low efficiency in parameter estimation, providing a high-quality data foundation for subsequent joint parameter estimation. Simultaneously, by optimizing signal amplitude and timing, it reduces the risk of device degradation due to over-excitation.

[0084] Example 4: Please refer to Figure 1 The specific method for step S3 is as follows:

[0085] S3.1: During the execution of the excitation signal, multi-source operating data is collected, time synchronization processing and event labeling processing are performed on the collected data, and data completion operations are performed on any missing data.

[0086] S3.2: Perform anomaly detection and credibility assessment on the collected data, remove data that does not meet the requirements, and generate a processed experimental dataset.

[0087] In this embodiment: This application further proposes to collect multi-source running data during the execution of the excitation signal, perform time synchronization processing and event labeling processing on the collected data, perform completion operation on missing data, perform anomaly detection and credibility assessment on the collected data, and remove data that does not meet the requirements to generate a processed experimental dataset.

[0088] Multi-source operational data acquisition refers to the synchronous acquisition of operating status, power curves, environmental parameters, and price signals from energy storage devices, adjustable loads, inverters, and market trading systems. This can be achieved using distributed data acquisition terminals and a unified clock source to ensure the comprehensiveness and synchronization of data sources. Time synchronization processing involves applying a unified timestamp to data generated by different devices. This can be achieved using network time protocols or hardware clock synchronization modules to eliminate time deviations caused by differences in device clocks. Event labeling processing involves marking time nodes for critical operations such as charging / discharging start / stop, load transfer triggering, and reactive power setting adjustments. This can be achieved using event-driven data loggers to facilitate subsequent analysis of the correlation between excitation signals and device responses. Data completion operation involves interpolating or model prediction to fill in missing or interrupted data segments. This can be achieved using linear interpolation, Kalman filtering, or long short-term memory neural networks to ensure the integrity of the data sequence. Anomaly detection involves identifying outliers that exceed the physical limits of the equipment or statistical regularities. This can be achieved using threshold-based judgment, isolated forest algorithms, or autoencoder models to eliminate noise interference. Credibility assessment refers to the quantitative scoring of data quality. Specifically, it can be achieved by using a multi-indicator fusion method based on data integrity, consistency, and time alignment to select data that meets the modeling requirements.

[0089] During the excitation signal execution phase, distributed acquisition terminals acquire real-time information on energy storage charging and discharging power, adjustable load operating status, inverter temperature, and market electricity prices. A unified clock source is used to synchronize data from each node at the millisecond level. To address potential data loss due to communication delays or equipment failures, a sliding window interpolation algorithm is used to locally fill in the power curve, ensuring the continuity of the time series. Energy storage start-up and shutdown events, load transfer commands, and reactive power setting adjustments are manually or automatically labeled to form a dataset with time-series labels. An isolated forest algorithm is used to detect unreasonable data points such as abnormal energy storage charging and discharging rates, sudden changes in load power, and inverter temperature exceeding limits. Cross-validation is performed using equipment operation logs to eliminate unreliable data segments. Finally, a high-reliability experimental dataset containing complete timestamps, event labels, and quality scores is generated, providing reliable input for subsequent parameter estimation.

[0090] Traditional methods typically collect data from a single data source and lack synchronization mechanisms, leading to temporal misalignment or loss of correlation among data from multiple devices. Existing technologies often address missing data by simply using the mean to fill in the gaps or by direct deletion, which can easily introduce bias or lose valuable information. Current anomaly detection methods mostly rely on fixed thresholds, making it difficult to adapt to dynamic changes in device operating conditions. This solution ensures data temporal consistency through multi-source synchronous acquisition and unified clock calibration, utilizes interpolation and prediction models to achieve high-precision data completion of missing data, and combines machine learning algorithms to dynamically identify complex anomaly patterns, significantly improving data quality and the reliability of subsequent parameter estimation.

[0091] This application effectively solves the difficulties in correlation analysis caused by asynchronous data acquisition from multiple devices, avoids the systematic errors introduced by traditional data completion methods, and overcomes the insufficient adaptability of fixed threshold detection in dynamic operating scenarios. The resulting experimental dataset features high time alignment, thorough outlier removal, and accurate missing information repair, providing a high-quality data foundation for subsequent joint parameter estimation and reducing the risk of parameter estimation bias due to data quality issues.

[0092] Example 5: Please refer to Figure 1 The specific method for step S4 is as follows:

[0093] S4.1: Establish a joint estimation model that includes energy storage parameters, load parameters, inverter parameters and price parameters. Introduce grid constraints, user constraints and market constraints as conditions in the model to form a parameter estimation framework that meets the operating conditions.

[0094] S4.2: Based on the joint estimation model, Bayesian inference or profile likelihood method is used to estimate the parameters, and the estimated values ​​of the parameters and their corresponding confidence intervals are output.

[0095] In this embodiment: This application further proposes to establish a joint estimation model that includes energy storage parameters, load parameters, inverter parameters, and price parameters; grid constraints, user constraints, and market constraints are introduced into the model as conditions to form a parameter estimation framework that meets the operating conditions; based on the joint estimation model, Bayesian inference method or profile likelihood method is used to estimate the parameters; the estimated values ​​of the parameters and their corresponding confidence intervals are output.

[0096] Joint estimation models integrate energy storage parameters, load parameters, inverter parameters, and price parameters into a unified mathematical framework for synchronous calculation. Specifically, they can be implemented using multivariate regression models or state-space models. By introducing grid constraints, user constraints, and market constraints as boundary conditions, the parameter estimation results are ensured to conform to actual operating conditions. Bayesian inference methods are based on Bayes' theorem, combining prior distributions and observed data to calculate the posterior probability distribution of parameters. Specifically, they can be implemented using Markov chain Monte Carlo sampling or variational inference algorithms, outputting parameter estimates and their confidence intervals in the form of probability distributions. Profile likelihood methods estimate the parameter range by fixing some parameters and maximizing the likelihood function of the remaining parameters. Specifically, they can be implemented using iterative optimization algorithms or gradient descent methods, determining the confidence interval by constructing a profile of the parameter likelihood function.

[0097] In the parameter estimation process, energy storage parameters, load parameters, inverter parameters, and price parameters are first integrated into a joint estimation model. Simultaneously, grid security constraints, user electricity consumption restrictions, and market trading rules are used as constraints on the model. For example, energy storage charging and discharging power limits, load adjustable range, and market price fluctuation range can be transformed into inequality constraints embedded in the model. Subsequently, a Bayesian inference method is used to calculate the posterior distribution of the parameters based on historical operating data and real-time collected data, through a probability distribution update mechanism, and outputs the parameter estimates and their confidence intervals. When there is a strong correlation between parameters, the profile likelihood method can be switched to obtain the feasible range of parameter estimation by gradually fixing some parameters and optimizing the remaining parameters.

[0098] Traditional methods typically estimate various parameters independently, neglecting the coupling relationships between parameters and operational constraints, leading to estimation results that deviate from actual operating conditions. This proposed solution, however, constructs a joint estimation model and introduces multidimensional constraints, ensuring that the parameter estimation process remains consistent with the actual operating boundaries. Furthermore, it employs probabilistic inference methods to quantify parameter uncertainty, providing an accurate input basis for subsequent scheduling optimization.

[0099] This application effectively solves the bias problem caused by neglecting constraints in the parameter estimation process, improving the reliability and practicality of the parameter estimation results. By outputting confidence intervals, it provides a quantitative basis for subsequent uncertainty set construction and robust optimization, thereby enhancing the adaptability and reliability of the virtual power plant scheduling strategy.

[0100] Example 6: Please refer to Figure 1 The specific method for step S5 is as follows:

[0101] S5.1: Combine parameters that are difficult to identify individually into equivalent parameters, and perform reparameterization on the parameter set;

[0102] S5.2: Generate a parameter uncertainty set based on the result of reparameterization. The uncertainty set is represented in the form of an ellipsoid set, box constraint, or polyhedron. The coverage of the uncertainty set is calibrated, and an upper and lower bound model of the parameters is established.

[0103] In this embodiment: This application further proposes a technical solution to combine parameters that are difficult to identify individually into equivalent parameters during the parameter estimation process, generate a parameter uncertainty set based on the estimation results, and adopt the form of an ellipsoid set, box constraint or polyhedron, and perform coverage calibration, while establishing an upper bound model and a lower bound model for the parameters.

[0104] Equivalent parameters refer to combining multiple parameters with strong correlations or indistinguishability into a single parameter with clear physical meaning through mathematical transformation. This can be achieved using linear combination, nonlinear mapping, or dimensionality reduction methods. For example, principal component analysis can be used to convert multiple energy storage degradation parameters into equivalent cycle life indices. The parameter uncertainty set refers to the set describing the possible range of parameter estimation results. Specifically, an ellipsoidal set can represent the confidence region of the parameter distribution, a box constraint can represent the upper and lower limits of independent parameter variation, and a polyhedron can represent the feasible region under linear inequality constraints. For example, an ellipsoidal set can be used to model the uncertainty of the energy storage capacity attenuation coefficient. Coverage calibration refers to adjusting the confidence level of the uncertainty set to match the statistical characteristics of the measured data. This can be achieved by adjusting the ellipsoid radius or the width of the box constraint boundary. For example, adjusting the uncertainty set of the price elasticity parameter based on historical data to achieve a coverage rate of 95%. Parameter upper and lower bound models refer to the boundary functions describing the maximum and minimum possible values ​​of the parameters, respectively. These can be constructed using extreme value theory or interval analysis methods. For example, the temperature dependence boundary of the inverter derating coefficient can be established using extreme value distributions.

[0105] When a strong correlation is detected between energy storage cycle efficiency and capacity decay coefficient during the parameter estimation stage, the two are combined into an equivalent energy storage health index, and parameter indiscernibility is eliminated through nonlinear transformation. An ellipsoidal uncertainty set is generated based on Bayesian inference results, and the ellipsoidal semi-axis length is adjusted to ensure the uncertainty set covers 90% of the measured data points. A box-constrained uncertainty set is constructed for the price elasticity parameter, and the boundary values ​​are dynamically adjusted based on user response data. Upper and lower bound models for the energy storage health index are established, and interval analysis is used to determine its fluctuation range under different temperature conditions. The uncertainty set and boundary models work together in the scheduling optimization process, providing a mathematical constraint basis for robust decision-making.

[0106] Existing methods typically treat unidentifiable parameters independently, leading to overly conservative or insufficient coverage issues in the construction of uncertainty sets. This proposed solution, however, effectively eliminates equivalence class interference between parameters through parameter combination and reparameterization. It employs multiple geometric uncertainties to adapt to different parameter characteristics and dynamically adjusts the confidence level using a coverage calibration mechanism, resulting in a more accurate representation of parameter uncertainty that closely matches the distribution characteristics of actual operational data.

[0107] This application solves the problem of inaccurate construction of uncertain sets caused by unidentifiable parameters in the prior art. By using equivalent parameter combination and coverage calibration mechanism, it improves the adaptability of uncertain sets to actual operating conditions while ensuring the mathematical rigor of uncertain sets, provides accurate parameter boundary constraints for multi-timescale scheduling, and effectively reduces the risk of scheduling strategy failure caused by parameter estimation bias.

[0108] Example 7: Please refer to Figure 1 The specific method for step S6 is as follows:

[0109] S6.1: In the scheduling framework of day-ahead, intraday and real-time multi-timescale, the uncertainty set and degradation model obtained in step S5 are introduced into the scheduling objectives and scheduling constraints to construct an optimization model that includes operational objective terms and information gain terms.

[0110] S6.2: Based on the optimization model, the solution is obtained by using the bibliometric optimization method or the chance constraint method, and the scheduling strategy that satisfies the constraint conditions is output.

[0111] In this embodiment: This application further proposes a dynamic resource optimization scheduling method for virtual power plants, which includes introducing uncertainty sets and degradation models into scheduling objectives and constraints in a multi-timescale scheduling framework of day-ahead, intraday, and real-time, and constructing an optimization model containing operational objective terms and information gain terms; based on the optimization model, the solution is performed using the split-bar optimization method or the chance constraint method, and the scheduling strategy that satisfies the constraint conditions is output.

[0112] A multi-timescale scheduling framework refers to dividing scheduling tasks into three phases: day-ahead, intraday, and real-time, for coordinated optimization. For example, the day-ahead phase can involve 24-hour planning, the intraday phase allows for hourly adjustments, and the real-time phase executes minute-level control. This division adapts to the dynamic changes in the power system across different time scales. The uncertainty set refers to the set of parameter value ranges generated through parameter estimation, which can be represented using ellipsoidal sets or box constraints. For example, limiting the energy storage capacity attenuation coefficient within a confidence interval. The information gain term refers to actively acquiring the effective data needed for parameter identification through the optimization model. For example, adding a penalty term for parameter estimation uncertainty to the scheduling objective. The fuzzy bar optimization method seeks the optimal solution while considering the uncertainty of parameter distribution, such as optimizing based on fuzzy sets constructed using Wasserstein distance.

[0113] During the day-ahead phase, when generating the basic scheduling plan, the set of parameter uncertainties is embedded as a constraint in the objective function; for example, upper and lower bounds of the capacity decay coefficient are introduced into the energy storage output constraint. During the intraday phase, when adjusting the plan based on real-time data, excitation signals are proactively designed using an information gain term to reduce parameter estimation errors. During the real-time phase, when executing scheduling instructions, a chance constraint method is used to handle the risks arising from parameter fluctuations; for example, a certain probability of constraint violation is allowed, but a preset confidence level must be met. Through multi-stage collaborative optimization, the scheduling strategy remains feasible under parameter uncertainty and changing operating conditions.

[0114] Traditional scheduling methods typically optimize only on a single time scale and fail to integrate parameter uncertainties with information gain mechanisms. Existing technologies often employ deterministic parameter assumptions, making them ineffective in addressing dynamic factors such as energy storage degradation and load elasticity variations. This proposed solution utilizes a multi-time-scale coupled optimization framework to dynamically link parameter identification with scheduling decisions, thereby enhancing the system's adaptability to complex operating environments.

[0115] This application effectively addresses the challenges of parameter uncertainty and dynamic coupling across multiple time scales in virtual power plant operation, enhancing the robustness and economy of the dispatching strategy. By introducing an information gain term to proactively reduce parameter estimation errors, it avoids dispatching commands becoming infeasible due to parameter deviations. Employing a combination of partial Bruker optimization and chance constraints, it improves the dispatching strategy's tolerance to parameter fluctuations while ensuring grid security constraints.

[0116] Example 8: Please refer to Figure 1 The specific method for step S7 is as follows:

[0117] S7.1: After the scheduling is completed, the opportunity constraint coverage is calibrated, and the parameter identification results are compared and evaluated with the scheduling scheme execution.

[0118] S7.2: Update the experimental design scheme and reparameterized model based on the comparative evaluation results, and generate and store scheduling process records and parameter update records.

[0119] In this embodiment: This application further proposes technical means to calibrate the opportunity constraint coverage after the scheduling is completed, compare and evaluate the parameter identification results with the scheduling scheme, update the experimental design scheme and reparameterized model based on the comparison and evaluation results, and generate and store scheduling process records and parameter update records.

[0120] Opportunity-constrained coverage calibration refers to verifying the consistency between the preset confidence level and the actual coverage probability by statistically analyzing the proportion of actual scheduling results that meet the opportunity constraints. Specifically, it can be achieved by using historical data backtracking analysis or Monte Carlo simulation methods to ensure the robustness of the scheduling strategy.

[0121] The comparative evaluation of parameter identification results and scheduling schemes refers to the correlation analysis between the parameter estimates and the actual operating data after scheduling execution. This can be achieved by residual analysis or error distribution testing methods, and is used to detect the deviation between the parameter model and the real system.

[0122] Updating the experimental design scheme refers to adjusting the generation strategy of the excitation signal based on the parameter estimation error distribution. Specifically, this can be achieved by dynamically adjusting the feasible range boundary or optimizing the target weight coefficient, in order to improve the effectiveness of subsequent experimental data.

[0123] Reparameterization model update refers to reconstructing the equivalent parameter combination form based on the calibration results of the parameter uncertainty set. Specifically, it can be achieved by principal component analysis or sensitivity ranking methods to eliminate the influence of unidentifiable parameters on the estimation process.

[0124] The generation of scheduling process records refers to the structured storage of scheduling strategies, constraints, and execution results in a time series. This can be achieved using time-series databases or blockchain notarization technology to support the traceability analysis of scheduling behavior.

[0125] After the scheduling execution phase, the coverage of the preset opportunity constraints is first verified. For example, the frequency of voltage over-limit events within the scheduling cycle is statistically analyzed to determine whether the confidence level reaches the expected threshold. If the coverage is insufficient, the constraint boundaries or confidence parameters are adjusted. Then, the parameter estimation results are compared with the actual response data of the scheduling scheme. For example, the effectiveness of the parameter model is evaluated by analyzing the difference between the predicted and measured values ​​of energy storage charging and discharging power. Based on the evaluation results, the amplitude, frequency, and other attributes of the excitation signal in the experimental design are dynamically optimized. For example, the intensity of electricity price incentives is increased during load response-sensitive periods. Simultaneously, the parameter combination is reconstructed. For example, the strongly correlated energy storage degradation coefficient and inverter efficiency parameters are merged into an equivalent health index. Finally, the complete scheduling logs, parameter update records, and verification reports are archived and stored, forming a closed-loop data chain.

[0126] Existing methods typically perform only a single parameter estimation before scheduling, lacking post-execution verification and model iteration mechanisms. For example, traditional scheduling strategies continue to use a fixed model when parameters are inaccurate, leading to accumulated biases. This solution, however, achieves dynamic feedback through coverage calibration and comparative evaluation. For instance, when a mismatch is detected between the inverter's thermal derating parameters and the actual temperature rise data, model reconstruction is immediately triggered, preventing the expansion of scheduling errors due to equipment aging.

[0127] This application effectively addresses the problem of dynamic mismatch between parameter models and real systems. For example, when energy storage capacity decay leads to overly aggressive scheduling strategies, closed-loop calibration can promptly correct parameter confidence intervals. It also improves the efficiency of data-driven model updates, such as quickly locating anomalous components in equivalent parameter combinations based on historical records. Furthermore, the complete process record provides a reliable basis for compliance auditing of scheduling strategies, such as verifying whether reactive power settings for specific time periods meet grid security constraints through source tracing analysis.

[0128] This application also includes a virtual power plant dynamic resource optimization scheduling system; please refer to [link / reference]. Figure 2 It includes a parameter modeling module, an incentive design module, a data acquisition module, a parameter estimation module, an uncertainty set module, a rolling scheduling module, and a verification and update module;

[0129] The parameter modeling module is used to perform structured modeling of energy storage degradation parameters, adjustable load utility parameters, inverter thermal derating parameters and price elasticity parameters, and to determine the set of identifiable parameters based on structured identifiability analysis. Parameters that are not identifiable or belong to equivalence classes are reparameterized to generate a set of candidate identifiable parameters.

[0130] The incentive design module is used to determine the feasible range of charging and discharging behavior, load shifting behavior and reactive power setting under the constraints of power grid security, user constraints and market constraints, construct the optimal experimental design model, solve the model based on Fisher's information criterion or mutual information criterion, and output the incentive scheme.

[0131] The data acquisition module is used to collect multi-source runtime data during the stimulus execution process, perform time synchronization, event labeling and missing data completion, and perform anomaly detection and credibility assessment on the data, remove data that does not meet the requirements, and generate experimental datasets.

[0132] The parameter estimation module is used to establish a joint estimation model that includes energy storage parameters, load parameters, inverter parameters and price parameters. The model introduces grid constraints, user constraints and market constraints as conditions, and calculates the parameter estimates and their corresponding confidence intervals based on Bayesian inference methods or profile likelihood methods.

[0133] The uncertainty set module is used to perform reparameterization processing on parameters that are difficult to identify individually, combine the parameters into equivalent parameters, and generate parameter uncertainty sets based on the estimation results. The parameter uncertainty sets are represented in the form of ellipsoidal sets, box constraints or polyhedra, and upper and lower bound models of the parameters are established.

[0134] The rolling scheduling module is used to introduce uncertainty sets and degradation models into scheduling objectives and constraints in a multi-timescale scheduling framework of day-ahead, intraday, and real-time, to construct an optimization model that includes operational objective items and information gain items, and solve the optimization model through the split-bar optimization method or the chance constraint method to output the scheduling strategy.

[0135] The verification and update module is used to calibrate the chance constraint coverage after the scheduling is completed, compare and evaluate the parameter identification results with the scheduling scheme, update the experimental design scheme and reparameterized model based on the comparison and evaluation results, and generate and store scheduling process records and parameter update records.

[0136] The contents not described in detail in this specification are existing technologies known to those skilled in the art.

[0137] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for dynamic resource optimization scheduling of a virtual power plant, characterized in that: The specific steps are as follows: S1: Perform structured modeling of energy storage degradation parameters, adjustable load utility parameters, inverter thermal derating parameters, and price elasticity parameters; determine the set of identifiable parameters based on structured identifiability analysis, and reparameterize parameters that are unidentifiable or have equivalence classes to obtain a set of candidate identifiable parameters; S2: Under the constraints of power grid security, user constraints and market constraints, excitation signals are generated for charging and discharging behavior, load shifting behavior and reactive power settings; S3: Construct the optimal experimental design model and solve the stimulus scheme using Fisher's information content or mutual information content criteria; S4: Establish a joint estimation model that includes energy storage parameters, load parameters, inverter parameters, and price parameters; introduce grid constraints, user constraints, and market constraints as conditions during the parameter estimation process; calculate the parameter estimates and corresponding confidence intervals using Bayesian inference or profile likelihood methods; S5: Combine parameters that are difficult to identify individually into equivalent parameters; generate parameter uncertainty sets based on the estimation results, wherein the uncertainty sets are in the form of ellipsoidal sets, box constraints or polyhedra, and perform coverage calibration; establish upper and lower bound models for the parameters; S6: In the multi-timescale scheduling framework of day-ahead, intraday, and real-time, the uncertainty set and degradation model obtained in step 5 are introduced into the scheduling constraints and objectives; an optimization model containing operational objectives and information gain terms is constructed, and the scheduling strategy is obtained by solving the bibliometric optimization method or the chance constraint method. S7: After scheduling execution, calibrate the opportunity constraint coverage; compare and evaluate the parameter identification results with the scheduling scheme; Update the experimental design and reparameterized model; generate and store records of the scheduling process and parameter updates.

2. The method for dynamic resource optimization scheduling of a virtual power plant according to claim 1, characterized in that: The specific method of step S1 is as follows: S1.1: Establish a mathematical relationship model between energy storage degradation parameters, adjustable load utility parameters, inverter thermal derating parameters, and price elasticity parameters and equipment operating status, external input and output variables. Based on the mathematical relationship model, use the structured identifiability analysis method to determine the set of identifiable parameters. S1.2: Reparameterize the detected unidentifiable parameters or parameters belonging to the equivalence class to obtain identifiable equivalent parameters, thereby generating a candidate set of identifiable parameters as input for subsequent experimental design and parameter estimation.

3. The method for dynamic resource optimization scheduling of a virtual power plant according to claim 2, characterized in that: The specific method of step S2 is as follows: S2.1: Under the constraints of power grid security, user constraints and market constraints, determine the feasible range of energy storage charging and discharging, load shifting and reactive power setting, generate excitation signals within the feasible range, and use the excitation signals as experimental inputs; S2.2: Construct an optimal experimental design model based on the excitation signal, introduce Fisher's information criterion or mutual information criterion into the model, solve the model, and obtain the optimal excitation scheme that satisfies the constraints.

4. The method for dynamic resource optimization scheduling of a virtual power plant according to claim 3, characterized in that: The specific method of step S3 is as follows: S3.1: During the execution of the excitation signal, multi-source operating data is collected, time synchronization processing and event labeling processing are performed on the collected data, and data completion operations are performed on any missing data. S3.2: Perform anomaly detection and credibility assessment on the collected data, remove data that does not meet the requirements, and generate a processed experimental dataset.

5. The method for dynamic resource optimization scheduling of a virtual power plant according to claim 4, characterized in that: The specific method of step S4 is as follows: S4.1: Establish a joint estimation model that includes energy storage parameters, load parameters, inverter parameters and price parameters. In the model, grid constraints, user constraints and market constraints are introduced as conditions to form a parameter estimation framework that meets the operating conditions. S4.2: Based on the joint estimation model, the parameters are estimated using Bayesian inference or profile likelihood methods, and the estimated values ​​of the parameters and their corresponding confidence intervals are output.

6. The method for dynamic resource optimization scheduling of a virtual power plant according to claim 5, characterized in that: The specific method of step S5 is as follows: S5.1: Combine parameters that are difficult to identify individually, combine the parameters into equivalent parameters, and perform reparameterization processing on the parameter set; S5.2: Generate a parameter uncertainty set based on the result of the reparameterization process. The uncertainty set is represented in the form of an ellipsoid set, box constraint, or polyhedron. The coverage of the uncertainty set is calibrated, and an upper and lower bound model of the parameters is established.

7. The method for dynamic resource optimization scheduling of a virtual power plant according to claim 6, characterized in that: The specific method of step S6 is as follows: S6.1: In the scheduling framework of day-ahead, intraday and real-time multi-timescale, the uncertainty set and degradation model obtained in step S5 are introduced into the scheduling objectives and scheduling constraints to construct an optimization model that includes operational objective terms and information gain terms. S6.2: Based on the optimization model, the solution is obtained by using the bibliometric optimization method or the chance constraint method, and the scheduling strategy that satisfies the constraint conditions is output.

8. The method for dynamic resource optimization scheduling of a virtual power plant according to claim 7, characterized in that: The specific method of step S7 is as follows: S7.1: After the scheduling is completed, the opportunity constraint coverage is calibrated, and the parameter identification results are compared and evaluated with the scheduling scheme execution. S7.2: Update the experimental design scheme and reparameterized model based on the comparative evaluation results, and generate and store the scheduling process record and parameter update record.

9. A virtual power plant dynamic resource optimization scheduling system, characterized in that: The virtual power plant dynamic resource optimization scheduling system is used to execute the virtual power plant dynamic resource optimization scheduling method described in any one of claims 1-8. The system includes a parameter modeling module, an incentive design module, a data acquisition module, a parameter estimation module, an uncertainty set module, a rolling scheduling module, and a verification and update module. The parameter modeling module is used to perform structured modeling of energy storage degradation parameters, adjustable load utility parameters, inverter thermal derating parameters and price elasticity parameters, and to determine the set of identifiable parameters based on structured identifiability analysis. Parameters that are not identifiable or belong to equivalence classes are reparameterized to generate a set of candidate identifiable parameters. The incentive design module is used to determine the feasible range of charging and discharging behavior, load shifting behavior and reactive power setting under the constraints of power grid security, user constraints and market constraints, construct the optimal experimental design model, solve the model based on Fisher's information criterion or mutual information criterion, and output the incentive scheme. The data acquisition module is used to collect multi-source runtime data during the stimulus execution process, perform time synchronization, event labeling and missing data completion on the data, and perform anomaly detection and credibility assessment on the data, remove data that does not meet the requirements, and generate an experimental dataset. The parameter estimation module is used to establish a joint estimation model that includes energy storage parameters, load parameters, inverter parameters and price parameters. The model introduces grid constraints, user constraints and market constraints as conditions, and calculates the parameter estimates and their corresponding confidence intervals based on Bayesian inference method or profile likelihood method. The uncertainty set module is used to perform reparameterization processing on parameters that are difficult to identify individually, combine the parameters into equivalent parameters, and generate a parameter uncertainty set based on the estimation results. The parameter uncertainty set is represented in the form of an ellipsoid set, box constraint, or polyhedron, and an upper bound model and a lower bound model for the parameters are established. The rolling scheduling module is used to introduce the uncertainty set and degradation model into the scheduling objectives and constraints in the multi-timescale scheduling framework of day-ahead, intraday and real-time, to construct an optimization model containing operational objective items and information gain items, and to solve the optimization model by the split-bar optimization method or the chance constraint method to output the scheduling strategy. The verification and update module is used to calibrate the chance constraint coverage after the scheduling is completed, compare and evaluate the parameter identification results with the scheduling scheme, update the experimental design scheme and reparameterized model based on the comparison and evaluation, and generate and store scheduling process records and parameter update records.

Images