Virtual power plant load optimization control method based on demand response

Abstract

The invention relates to the technical field of energy internet control, and discloses a virtual power plant load optimization control method based on demand response. The method comprises the steps that real-time operation data and a power grid instruction are automatically obtained, dynamic behavior characteristics such as the load change rate and the fluctuation period are mined after data cleaning, and a load control strategy draft is generated according to the dynamic behavior characteristics. And then configuring control node parameters and collecting performance data, obtaining a current dynamic operation constraint of the system through simulation calculation, carrying out load distribution optimization solution on the condition, and finally generating a control instruction sequence and carrying out deployment execution. According to the method, the matching degree of the control strategy and the actual load response capability is improved through load dynamic behavior characteristic analysis, the feasibility of the optimization result is ensured by utilizing the dynamic constraint defined by simulation feedback, and more accurate and safer virtual power plant load regulation is realized.

Description

Technical Field

[0001] This invention relates to the field of energy internet control technology, specifically to a demand-response-based virtual power plant load optimization control method. Background Technology

[0002] Virtual power plants, as an important form of integrating distributed resources to participate in grid operation, rely on load optimization control as a key to realizing demand response value. Existing technologies typically rely on the static attributes of the load for control. Upon receiving a grid demand response command, the system generates a load control strategy based on preset rules. This static tag-based control mode constitutes the current mainstream solution.

[0003] This static control model has flaws. It ignores the dynamic behavior of the load itself, treating each load unit as a simple switch or a constant power unit. In reality, different types of loads, and even different individuals of the same type, have significantly different inherent patterns of power variation and response potential. For example, some loads can quickly reach their target power but are difficult to maintain, while others are suitable for slow, long-term, stable regulation. The static model cannot capture these differences in dynamic behavior, leading to a mismatch between the control strategy and the actual response capability of the load. This can easily cause problems such as inadequate execution of control commands, large fluctuations in the regulation process, or negative impacts on the user's normal electricity experience.

[0004] Existing optimization methods often rely on pre-defined, fixed system operating constraints when formulating control strategies. These constraints are typically based on theoretical limits of the power grid architecture or conservative empirical estimates, failing to correlate with the actual operating state of the current virtual power plant and the dynamic impact of the control strategy to be implemented. Using fixed constraints for optimization may result in overly conservative results, failing to fully exploit the regulation potential of the virtual power plant; or the results may be overly idealistic, potentially approaching or even exceeding the safe operating boundaries of the local system during actual implementation, leading to potential risks. This disconnect between the generation of control strategies and the real-time operational safety of the power grid limits the accuracy and reliability of the virtual power plant's control effects. Summary of the Invention

[0005] The purpose of this invention is to provide a demand-response-based virtual power plant load optimization control method to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a demand-response-based virtual power plant load optimization control method, the method comprising: The data acquisition and control system automatically acquires the real-time operation dataset of the virtual power plant and the grid demand response instructions. The real-time operation dataset includes load power sequence and adjustable load resource information, while the grid demand response instructions include response time interval and load adjustment target value. Perform data cleaning and format standardization operations on the real-time running dataset to remove invalid data and unify data scale, and generate a preprocessed dataset. Mining load behavior characteristics from the preprocessed dataset, including load change rate and fluctuation cycle characteristics; By integrating load behavior characteristics and grid demand response commands, load resources are adapted, and a draft load control strategy is output. Configure the control node parameters of the virtual power plant according to the draft load control strategy, and collect the performance attribute data of the control nodes; Based on performance attribute data and draft load control strategies, simulation tools are used to calculate system operating constraints and define power variation boundaries and deviation tolerances. By using system operating constraints as optimization conditions, load allocation is solved to obtain optimized control commands; The control commands are analyzed and optimized, the control action time series is constructed, and the control action time series is deployed to the virtual power plant control platform to perform load regulation.

[0007] Preferably, mining load behavior features from the preprocessed dataset includes: A time-dimensional analysis was performed on the preprocessed dataset to extract the long-term trend component and short-term fluctuation component of the load power sequence. Statistical indicators for calculating short-term fluctuation components, including standard deviation and extreme value distribution; Identify the periodic patterns of the load power sequence and use spectral analysis to determine the dominant frequency component; By integrating statistical indicators of long-term trend components, short-term fluctuation components, and periodic patterns, a set of load behavior characteristics is formed.

[0008] Preferably, the draft load control strategy for load resource adaptation includes: Establish a load resource rule base, which contains mapping relationships between various load types and regulation capabilities; The similarity between load behavior characteristics and load adjustment target values ​​is calculated to generate a matching score; Candidate load resources in the adjustable load resource information are filtered based on the matching score; Perform capacity verification on candidate load resources to ensure that the response time interval requirements are met; Combine candidate load resources and their regulation capabilities to generate a draft load control strategy.

[0009] Preferably, the system operational constraints calculated using simulation tools include: Load control node parameters and performance attribute data into the simulation environment; Set up simulation scenarios to simulate the system response under different load adjustment ranges; Collect simulation output data, including voltage deviation and frequency variation data; Analyze voltage deviation and frequency variation data to determine the critical conditions for stable system operation; The power variation boundary and deviation tolerance are derived based on the critical conditions.

[0010] Preferably, the load distribution solution is performed to obtain the optimal control command, including: A load optimization model is constructed, with the objective function being to minimize the grid deviation. Inject power variation boundary and deviation tolerance as constraints into the load optimization model; The discretized response time interval is divided into multiple time segments; For each time segment, load adjustment amounts are allocated, and the feasible solution space is calculated based on adjustable load resource information. An iterative search algorithm is used to find the optimal solution in the feasible solution space and output optimized control instructions.

[0011] Preferably, before constructing the load optimization model, a load forecasting model is pre-built, including: Historical load records associated with load behavior characteristics and performance attribute data are extracted from the historical database. These historical load records include sample load command sequences, sample adjustment result sequences, deviation change sequences, and power change sequences. Historical load records are analyzed over time series, and the data are divided into multiple subsequences at fixed time intervals, with each subsequence serving as a training sample. For each training sample, the first and second derivatives of the power change sequence are calculated to extract the power change gradient features. At the same time, the mean and variance of the deviation change sequence are calculated to extract the deviation fluctuation features. By combining the power change gradient features and the deviation fluctuation features, a multi-dimensional feature vector is generated; A convolutional neural network model is trained using multidimensional feature vectors as input data. The convolutional neural network model contains convolutional layers, pooling layers, and fully connected layers. The convolutional layers are used to extract local temporal patterns, the pooling layers are used for dimensionality reduction, and the fully connected layers are used to output the predicted load value. The parameters of the convolutional neural network model are iteratively updated using the gradient descent algorithm to minimize the mean square error between the predicted load value and the actual load value. When the training error reaches a preset threshold, training stops, and the load prediction model is obtained.

[0012] Preferably, the parsed and optimized control instructions include: Decompose and optimize the load adjustment time and deviation control requirements in the control instructions; Compare the load adjustment time and the time limit of the deviation control requirement, and take the minimum value as the control cycle benchmark; Define the cost weight configuration, which includes economic weight, stability weight, and reliability weight; Construct a comprehensive evaluation function based on cost weighting; Retrieve control cases with similar scenarios from historical operation records. Each control case includes sample control parameters and sample execution cost. The sample control parameters include load adjustment rate, adjustment duration, and stability maintenance interval. The sample execution cost includes energy consumption, equipment loss degree, and grid deviation integral. The comprehensive evaluation function is used to evaluate the control case and select the optimal control parameters. By combining the control cycle benchmark with the optimal control parameters, a control action time series is generated.

[0013] Preferably, the automatic acquisition of real-time operational datasets of the virtual power plant and grid demand response commands through the data acquisition and control system includes: Deploy a distributed sensor network to continuously collect operational data from the load units; Receive demand response signals issued by the power grid center through the communication interface; Verify the integrity and consistency of operational data and demand response signals, and filter outliers; Aggregate operational data and demand response signals to form a standardized data flow.

[0014] Preferably, the construction of the comprehensive evaluation function based on cost weight configuration is performed according to the following steps: Define evaluation dimensions, which include economic dimension, stability dimension and reliability dimension; Each evaluation dimension is assigned a weighting coefficient, which is dynamically adjusted according to the priority of the power grid demand response command. From an economic perspective, quantify the changes in energy costs associated with load adjustment; Regarding the stability dimension, the impact of load fluctuations on system frequency is quantified; For reliability, the expected success rate of load regulation is quantified; The quantized values ​​of different dimensions are normalized to eliminate the difference in dimensions. The normalized quantized values ​​are weighted and summed using weighting coefficients to generate a comprehensive evaluation function. The comprehensive evaluation function is calibrated using historical operational data to optimize the weight coefficient values.

[0015] Preferably, the analysis of voltage deviation and frequency variation data to determine the critical conditions for stable system operation is performed according to the following steps: Extract the voltage deviation sequence and frequency variation sequence from the simulation output data; Set the voltage deviation threshold and frequency change threshold respectively; Traverse the voltage deviation sequence to identify extreme points that exceed the voltage deviation threshold and their duration; Traverse the frequency change sequence to identify extreme points that exceed the frequency change threshold and their duration; Calculate the system stability margin by associating the voltage extreme points and frequency extreme points at the same time section; Using the state where the system stability margin approaches zero as the critical condition, record the corresponding load adjustment range combination under this state; The boundary parameters for stable system operation are derived by combining the load adjustment range.

[0016] Compared with the prior art, the beneficial effects of the present invention are: By mining behavioral characteristics such as load change rate and fluctuation cycle from preprocessed operational data and integrating these dynamic features with grid demand response commands, the generation of load control strategies has shifted from being based on static attributes to being based on dynamic behavioral patterns. Load change rate characteristics can identify which load resources have rapid response capabilities and are suitable for handling steep power ramp-up demands; fluctuation cycle characteristics help determine whether the load is suitable for short-term impulsive adjustments or long-term continuous adjustments. This matching based on behavioral characteristics allows control strategies to better align with the inherent response patterns of various loads, improving the fit between load resources and the dynamic demands of the grid. This enhances the accuracy and flexibility of load control, reduces command execution deviations, lowers system power fluctuations, and improves user power comfort.

[0017] By configuring preliminary strategies to control nodes and collecting their performance attribute data, and then using simulation tools to calculate specific power change boundaries and deviation tolerances under the current state, a dynamic and high-fidelity optimization solution environment is constructed. This process ensures that the constraints followed by the optimization algorithm are no longer static and conservative, but truly reflect the instantaneous carrying capacity and safety margin of the virtual power plant system when executing a specific control strategy. Using these dynamic constraints as conditions for load allocation solutions, the resulting optimized control commands can maximize the utilization of the system's currently acceptable regulation capacity while ensuring the feasibility and safety of the commands. This achieves safe tapping of the virtual power plant's regulation capacity, avoids the conservative or risky problems that may arise from optimization results deviating from actual operating conditions, and improves the reliability and execution success rate of the control strategy. Attached Figure Description

[0018] Figure 1 This is a schematic diagram illustrating the working principle of the demand-response-based virtual power plant load optimization control method described in this invention. Figure 2 A flowchart for identifying load behavior characteristics; Figure 3 A flowchart for calculating the system's operational constraints; Figure 4 This is a frequency deviation correlation diagram; Figure 5 This is a power tracking diagram showing the time series of load regulation and control actions in a virtual power plant. Detailed Implementation

[0019] 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.

[0020] Please see Figure 1 This invention provides a load optimization control method for a virtual power plant based on demand response. The method includes: a data acquisition and control system automatically acquiring real-time operating datasets of the virtual power plant and grid demand response commands. The real-time operating datasets contain load power sequences and adjustable load resource information, while the grid demand response commands include response time intervals and load adjustment target values. After acquiring the data, data cleaning and format standardization operations are performed on the real-time operating datasets to remove invalid data and unify data scales, generating a preprocessed dataset. Load behavior characteristics, including load change rate and fluctuation cycle characteristics, are mined from the preprocessed dataset. The load behavior characteristics and grid demand response commands are fused to perform load resource adaptation and output a draft load control strategy. The control node parameters of the virtual power plant are configured according to the draft load control strategy, and performance attribute data of the control nodes are collected. Based on the performance attribute data and the draft load control strategy, system operating constraints are calculated using simulation tools, and power change boundaries and deviation tolerances are defined. The system operating constraints are used as optimization conditions to solve for load allocation and obtain optimized control commands. The optimized control commands are parsed to construct a control action time series, and the control action time series is deployed to the virtual power plant control platform to perform load regulation.

[0021] Example 1: See Figure 2In practical implementation, the data acquisition stage of the demand response-based virtual power plant load optimization control method continuously collects operational data from load units through the deployment of a distributed sensor network. This network covers all key nodes of the virtual power plant, including generation units, load units, and energy storage units. Sensor types include smart meters and power sensors, monitoring load power, voltage, and current parameters in real time. The system receives demand response signals from the power grid center via a communication interface using standard protocols such as IEC 61850. These signals are transmitted in digital message format, containing the response time interval and load adjustment target value. Verification of the integrity and consistency of operational data and demand response signals employs cyclic redundancy check (CRC) algorithms and data packet sequence number checks, filtering outomas such as power readings exceeding reasonable ranges or data points with incorrect timestamps. Finally, operational data and demand response signals are aggregated to form a standardized data stream. This aggregation process is based on timestamp alignment and data format conversion. The standardized data stream is stored in structured database tables or time-series files.

[0022] It is understandable that after data acquisition, the data preprocessing stage begins. This involves performing data cleaning on the real-time dataset to remove invalid data such as missing values ​​or duplicate records. Automated scripts are used to identify and delete data rows that do not meet the criteria. Format standardization operations unify data scales, including unit conversion and numerical normalization. For example, power values ​​are standardized to megawatts, and timestamps are converted to the standard ISO format. In practice, mining load behavior characteristics from the preprocessed dataset begins with time-dimensional analysis. This analysis uses a sliding window method to segment the load power sequence into multiple time segments. The long-term trend component of the load power sequence is extracted using moving averages or exponential smoothing, reflecting the macroscopic direction of load changes. Short-term fluctuation components are extracted using differential methods or high-pass filters, capturing instantaneous load changes.

[0023] In practice, the first step in time-dimensional analysis is data resampling to ensure uniform time intervals. Resampling methods include upsampling or downsampling to suit the analytical needs. Long-term trend components are extracted using a moving average algorithm with an adjustable window size, dynamically set according to the speed of load changes. In practice, short-term fluctuation components are separated by subtracting the trend components to obtain a residual sequence, which represents the fluctuation portion. Standard deviation is calculated based on the square root of the variance of the residual sequence, and extreme value distributions are identified using percentiles or box plots. Periodic pattern recognition first calculates the autocorrelation function of the power sequence to initially detect the period, then applies spectral analysis to precisely extract frequency components.

[0024] Example 2: In practical implementation, the load resource adaptation process begins with establishing a load resource rule base. This rule base is a structured database that stores the mapping relationships between various load types and their regulation capabilities. Load types include specific categories such as industrial loads, commercial loads, and residential loads. Industrial loads might correspond to large motors or production lines, commercial loads cover air conditioning and lighting systems in shopping malls, and residential loads involve household appliances. Regulation capabilities quantitatively describe the adjustable power range, regulation rate, and duration for each load type within a specific time period. The mapping relationships exist in the form of rule entries; for example, a rule might record "Industrial load type A, maximum power reduction 500 kW, regulation rate 100 kW / min, minimum continuous response time 30 minutes." The load resource rule base is built and maintained by continuously inputting historical operating data and equipment nameplate parameters to ensure the accuracy and timeliness of the mapping relationships.

[0025] In practice, similarity calculations are performed between load behavior characteristics and load adjustment target values. Load behavior characteristics are vectors mined from preprocessed data, containing features such as average power, fluctuation period, and rate of change. The load adjustment target value comes from the grid demand response command and is a specific power numerical target. The similarity calculation aims to assess the degree of matching between the current load behavior pattern and the target state to be achieved. One specific calculation method is to use a weighted Euclidean distance or cosine similarity algorithm. For example, a similarity function can be defined that comprehensively compares the various dimensions of the load behavior characteristics with the expected characteristics of the load adjustment target value. The matching score is a numerical result; a higher matching score indicates that the load behavior characteristics are more likely to approach the load adjustment target value through adjustment. The matching score calculation process considers the time urgency of the load adjustment target value; for commands with short response time intervals, the matching degree of the load change rate feature is given more weight.

[0026] In practice, candidate load resources are selected from the adjustable load resource information based on matching scores. This adjustable load resource information is a list containing all technically adjustable load resources within the virtual power plant and their real-time status. The selection process sets a matching score threshold; load resources with matching scores higher than this threshold are initially selected as candidate load resources. The matching score threshold can be a fixed value or dynamically adjusted based on the ratio of the current total load adjustment target value to the total available resources. For example, when the adjustment task is heavy and resources are relatively scarce, the matching score threshold can be appropriately lowered to broaden the candidate range. The selection operation ensures that the inherent characteristics of the selected candidate load resources have a high degree of fit with the current adjustment task.

[0027] In practice, capacity verification is performed on candidate load resources to ensure that their actual adjustable capacity at the current moment meets the response time interval requirements. The verification includes two aspects: first, verifying whether the available adjustable capacity of the candidate load resource at the current moment is greater than the load adjustment amount it needs to share; and second, verifying whether the time required for the candidate load resource to complete the adjustment of that amount is within the allowable range of the response time interval. Capacity verification requires querying the adjustment rate of this type of load in the load resource rule base and calculating it in conjunction with the current status of the load in real-time operational data. For example, if a candidate load resource needs to undertake a 100 kW load reduction task with an adjustment rate of 20 kW / min, then completing the adjustment will take at least 5 minutes. If the response time interval requirement is that the adjustment must take effect within 10 minutes and last for 30 minutes, then the resource passes the capacity verification. If a candidate load resource cannot meet the capacity requirements, it is removed from the candidate list. Capacity verification is a crucial step in ensuring the feasibility of the load control strategy.

[0028] In practice, candidate load resources and their regulation capabilities are combined to generate a draft load control strategy, which serves as a preliminary action plan. The combination process needs to consider the coordination between multiple candidate load resources, aiming to optimize resource utilization while meeting the overall load adjustment target. A simple combination method is to allocate regulation tasks sequentially according to matching scores until the total adjustment amount is met. More complex combinations may consider factors such as the geographical distribution of load resources and regulation costs. The generated draft load control strategy clearly lists each candidate load resource participating in regulation, the regulation amount each candidate load resource needs to undertake, the start time of regulation, and the expected regulation curve. The draft load control strategy is a text or structured data object for further optimization and verification in subsequent stages.

[0029] In implementation, evaluation dimensions are defined, including economic, stability, and reliability dimensions. Each dimension is assigned a weighting coefficient, which is dynamically adjusted based on the priority of the grid demand response command. The economic dimension focuses on cost changes resulting from load regulation, such as electricity price differences or equipment wear and tear costs. The stability dimension focuses on the impact of the regulation process on grid stability indicators such as frequency and voltage. The reliability dimension focuses on the expected probability of successful execution of the load regulation command. The weighting coefficients determine the relative importance of each dimension in the subsequent comprehensive evaluation. For example, when the grid demand response command explicitly requires prioritizing system safety, the stability dimension's weighting coefficient is set to the highest; conversely, when the command emphasizes economic optimization, the economic dimension's weighting coefficient is highest. The dynamic adjustment mechanism of the weighting coefficients is typically based on semantic analysis of the command content or a pre-defined rule base.

[0030] In practical implementation, regarding the economic dimension, the change in energy costs of load regulation is quantified. This quantification process is based on a precise mathematical model. Changes in energy costs include not only changes in electricity expenses due to changes in electricity consumption, but may also include ancillary service fees, carbon emission costs, etc. Specific calculations require input parameters such as real-time electricity prices, load regulation amount, and regulation duration. Regarding the stability dimension, the impact of load fluctuations on system frequency is quantified. This impact can be obtained through theoretical calculations or simulations. For example, the integral of frequency deviation over time, or the maximum frequency deviation caused by a step change in load, can be used as a quantification indicator. Regarding the reliability dimension, the expected success rate of load regulation is quantified. This expected success rate can be comprehensively assessed based on historical response records of load resources, current health status, and the reliability of communication channels. One quantification method is to use historical data to statistically analyze the success probability of similar loads under similar operating conditions.

[0031] In practical implementation, the quantified values ​​of different dimensions are normalized to eliminate dimensional differences. Normalization allows the three different indicators—economic efficiency, stability, and reliability—to be compared and calculated on the same scale. Commonly used normalization methods include min-max normalization or Z-score standardization. For example, min-max normalization linearly transforms the original quantified values ​​of each dimension to the [0,1] interval. Weighting coefficients are then applied to the normalized quantified values ​​for a weighted sum, generating a comprehensive evaluation function. This comprehensive evaluation function maps a draft load control strategy to a single score, which comprehensively reflects the draft's overall performance in terms of economic efficiency, stability, and reliability. Its mathematical expression can be represented as: in: Represents the overall evaluation score; , , These represent the weighting coefficients assigned to the economic, stability, and reliability dimensions, respectively. , , These represent the normalized results of the quantitative values ​​for the economic, stability, and reliability dimensions, respectively. Historical operational data is used to calibrate the comprehensive evaluation function and optimize the weight coefficients. The calibration process compares the predictive performance of the comprehensive evaluation function with the actual execution results. Regression analysis and other methods can be used to adjust the weight coefficients, ensuring that the strategy proposal with the highest comprehensive evaluation score actually produces better overall results in practice.

[0032] Example 3: See Figure 3In practical implementation, when calculating system operating constraints using simulation tools, the first step is to load control node parameters and performance attribute data into the simulation environment. Control node parameters originate from the load control strategy draft and specifically include the identifiers of the load resources involved in regulation, initial power setpoints, maximum and minimum power limits, and communication delay parameters. Performance attribute data is obtained in real-time from the virtual power plant monitoring system and includes the output characteristics of generator sets, the impedance parameters of transmission lines, the transformer ratio and capacity, and the setting information of protection devices. The simulation environment is typically built using specialized power system analysis software. The loading process maps the control node parameters and performance attribute data to corresponding components in the simulation model through predefined interface scripts. For example, it replaces the static load model parameters of the load nodes with the adjustable load model parameters in the control node parameters, and updates the governor model parameters of the generator sets to the current values ​​in the performance attribute data.

[0033] In practical implementation, simulation scenarios are set up to simulate the system response under different load adjustment ranges. The simulation scenarios are constructed based on the load adjustment target value and response time interval in the power grid demand response command. The load adjustment range is applied in the simulation in a step-like or continuously ramped manner, for example, starting from 50% of the load adjustment target value and gradually increasing to 150% in 10% increments to cover possible under-adjustment and over-adjustment situations. Each adjustment range corresponds to an independent simulation case, which clearly specifies the total load change, the rate of change, and the start and end times of the change. The simulation tool performs calculations according to the preset time step, solving the differential algebraic equations of the system and simulating the electromechanical transient processes inside the virtual power plant. The system response is recorded in detail, focusing on the change trajectories of key physical quantities such as node voltage, system frequency, and line power. When setting up simulation scenarios, the worst-case scenario needs to be considered, such as simulating a tie line tripping simultaneously with a sudden load change, to test the system's anti-interference capability.

[0034] In practice, simulation output data is collected, including voltage deviation sequences and frequency change sequences. The voltage deviation sequence is the set of voltage deviations from the rated voltage at each monitoring node arranged chronologically during the simulation. The frequency change sequence is the set of deviations from the system center frequency (50Hz or 60Hz) arranged chronologically. Data acquisition is automatically completed by the simulation tool's data recording module, with the recording interval consistent with the simulation step size. For the voltage deviation sequence, the focus is typically on voltage changes at grid hub nodes and load-concentrated areas; for the frequency change sequence, the dynamic process of the overall system frequency is recorded. The simulation output data is stored in the form of a matrix or time series array, with each row representing a time point and each column representing a monitored variable. In addition to the voltage deviation sequence and frequency change sequence, the simulation output data typically includes auxiliary analysis data such as generator power angle and line load rate.

[0035] In practical implementation, the voltage deviation sequence and frequency change sequence are analyzed to determine the critical conditions for stable system operation. The analysis process first requires setting voltage deviation thresholds and frequency change thresholds respectively. The voltage deviation threshold is set according to the power system safety operation regulations, for example, stipulating that the voltage deviation should not exceed ±5% of the rated voltage. The frequency change threshold is also based on regulations, for example, stipulating that the frequency deviation should not exceed ±0.2Hz. These thresholds are hard indicators for judging whether the system is within a stable and safe range. Subsequently, the voltage deviation sequence is traversed to identify extreme points exceeding the voltage deviation threshold and their durations. Extreme points are local maximum or minimum points in the voltage deviation sequence, and the duration refers to the length of time during which the voltage deviation continuously exceeds the threshold. Similarly, the frequency change sequence is traversed to identify extreme points exceeding the frequency change threshold and their durations. The voltage and frequency extreme points at the same time point are correlated to calculate the system stability margin. The system stability margin is a comprehensive indicator used to quantify the distance between the current system state and the instability boundary. A method for calculating system stability margin is described below. The method is as follows: in: It represents the stability margin of the system. The closer its value is to 1, the more stable the system is. The closer it is to 0, the closer it is to the critical instability state. This represents the maximum absolute value of the voltage deviation sequence within the associated time span. This represents the preset voltage deviation threshold. It represents the maximum absolute value of the frequency change sequence within the same time span. This represents a preset frequency variation threshold. It is used for system stability margin. The state approaching zero is used as the critical condition. The corresponding load adjustment range combination in this state is recorded. The load adjustment range combination refers to the configuration of various load resources that lead the system to a critical state. Based on the load adjustment range combination, the boundary parameters for stable system operation are deduced. These boundary parameters are specified as power change boundaries and deviation tolerance. The power change boundaries define the maximum allowable power change of various loads per unit time, and the deviation tolerance defines the limit values ​​of instantaneous voltage and frequency deviations allowed during load adjustment. It can be understood that the determination of the critical condition is conservative, leaving a safety margin for actual operation. It can also be understood that the system stability margin... The calculation method can be adjusted according to system characteristics, for example, by adding weighting factors to highlight the importance of voltage stability or frequency stability.

[0036] In practice, after extracting the voltage deviation and frequency variation sequences from the simulation output data, analysis can be performed using specialized scripts or visualization tools. Identifying extreme points by traversing the sequences typically employs algorithms that find the zero-crossing points of the numerical derivative. Correlating voltage and frequency extreme points at the same time point means aligning the time axis to identify moments when voltage and frequency deteriorate simultaneously or sequentially. The calculated system stability margin... This is an important evaluation metric for each simulation scenario. It's crucial for determining the system stability margin corresponding to a specific combination of load adjustments. When the load falls below a preset safety threshold (e.g., 0.1), the system is considered to be in a critical operating state. All operating parameters corresponding to this critical condition are recorded, especially the total load adjustment, adjustment rate, and the contribution ratio of each load resource. The process of back-deriving boundary parameters involves reducing these critical state operating parameter values ​​by a certain safety factor to determine the power change boundary and deviation tolerance for stable system operation. For example, if the critical state occurs when the total load increases by 20 MW instantaneously, the power change boundary might be set at 15 MW, and such adjustments must be completed slowly over 30 seconds rather than instantaneously. Deviation tolerance might be set to a voltage deviation of no more than 4% and a frequency deviation of no more than 0.15 Hz, which is more stringent than the critical threshold.

[0037] See Figure 4 The graph shows the load adjustment range on the horizontal axis, the voltage deviation on the left vertical axis, and the frequency deviation on the right vertical axis. The curves illustrate the changing trends of both as the load adjustment range increases. This graph contains voltage / frequency deviation data collected after loading control node parameters and setting different load adjustment range simulation scenarios, intuitively demonstrating the correlation between load adjustment range and grid stability indicators. This graph is a crucial basis for determining the critical conditions for stable system operation. By identifying the adjustment range where the voltage / frequency deviation exceeds the safety threshold, the power change boundary and deviation tolerance can be inferred, providing dynamic safety constraints for subsequent load optimization solutions. It is a core supporting link in ensuring the safety and feasibility of virtual power plant regulation.

[0038] Example 4: In practical implementation, to obtain the optimal control command by solving the load allocation problem, it is first necessary to construct a load optimization model. The load optimization model is a mathematical programming model, and its objective function is set to minimize the grid deviation, that is, minimize the difference between the total load of the virtual power plant and the load adjustment target value in the grid demand response command. The decision variables of the load optimization model are the power adjustment amount of each adjustable load resource in the discretized time segment. The power change boundary and deviation tolerance are injected into the load optimization model as constraints. The power change boundary restricts the power change of each load resource in the form of inequality constraints to not exceed its allowable upper limit in a unit time. The deviation tolerance is used as a global constraint, requiring that the maximum voltage or frequency deviation in the entire optimization cycle does not exceed the safety threshold calculated by simulation. The discretized response time interval is divided into multiple time segments. For example, a 60-minute response time interval is divided into 12 5-minute time segments, and each time segment becomes an independent optimization period. For each time segment, load adjustment amounts are allocated. A feasible solution space is calculated based on adjustable load resource information. This information provides the upper and lower limits of the adjustment capacity, adjustment rate, and cost coefficient for each load resource. The feasible solution space is the set of all load allocation schemes that satisfy the physical constraints of the load resources themselves and the power change boundary constraints. An iterative search algorithm is used to find the optimal solution in the feasible solution space and outputs optimized control commands. Iterative search algorithms, such as genetic algorithms or particle swarm optimization algorithms, simulate biological evolution or swarm intelligence behavior, evaluating a large number of candidate solutions in parallel within the solution space, gradually approximating the optimal load allocation scheme that minimizes the objective function. The optimized control commands specify the power setpoint that each load resource needs to execute in each time segment.

[0039] In practical implementation, before constructing the load optimization model, a pre-built load forecasting model is established. This model extracts historical load records associated with load behavior characteristics and performance attributes from a historical database. These historical load records include sample load command sequences, sample adjustment result sequences, deviation change sequences, and power change sequences. The sample load command sequence is a chronological sequence of load regulation commands issued to the virtual power plant in history. The sample adjustment result sequence is the actual load change sequence executed by the virtual power plant. The deviation change sequence is the difference between the sample load command sequence and the sample adjustment result sequence. The power change sequence is the rate of change of load power over time. Time series analysis is performed on the historical load records, dividing them into multiple subsequences at fixed time intervals. Each subsequence serves as a training sample. The fixed time interval is typically consistent with the length of the time segment used in future optimization, such as 5 minutes or 15 minutes. For each training sample, the first and second derivatives of the power change sequence are calculated to extract power change gradient features. The first derivative reflects the trend of power change, and the second derivative reflects the acceleration of the change trend. Simultaneously, the mean and variance of the deviation change sequence are calculated to extract deviation fluctuation features. The mean represents the average deviation level, and the variance represents the degree of deviation fluctuation. The power change gradient features and deviation fluctuation features are combined to generate a multi-dimensional feature vector, which serves as input describing the dynamic characteristics of the load. Using this multi-dimensional feature vector as input data, a convolutional neural network (CNN) model is trained. The CNN model includes convolutional layers, pooling layers, and fully connected layers. The convolutional layers extract local temporal patterns by sliding the convolutional kernel along the time dimension. The pooling layers downsample the output of the convolutional layers to achieve dimensionality reduction. The fully connected layers integrate all features and output the predicted load value. The parameters of the CNN model are iteratively updated using the gradient descent algorithm to minimize the mean squared error between the predicted and actual load values. Training stops when the training error reaches a preset threshold, resulting in a successfully trained load prediction model.

[0040] In practical implementation, the load optimization model can be formalized as a constrained optimization problem. The objective function aims to minimize the total deviation over the entire response time interval. A common form of the objective function is to minimize the sum of squares of the deviations in each time segment. Its mathematical expression can be represented as: in: This represents the value of the objective function that needs to be minimized. This represents the total number of time segments obtained after discretization within the response time interval. It is the index of the time segment, from 1 to... . Represents a time segment Within the virtual power plant, the total power prediction value of all loads is output by the load prediction model and is a function of the decision variables. Represents a time segment Within this context, the target load adjustment value is the one required by the grid demand response command. Constraints include: the power value of each adjustable load resource in each time segment must not exceed its maximum or minimum technical limit; the power change of the same load resource between adjacent time segments must not exceed the rate limit specified by the power change boundary; and the voltage deviation and system frequency deviation of key system nodes derived from the optimization results must not exceed the deviation tolerance. Refer to Table 1; adjustable load resource information is typically stored and retrieved in Table 1 format.

[0041] Table 1: Adjustable Load Resource Information Table In some embodiments, the load optimization model may consider network losses and incorporate power flow constraints into the model, at which point the problem becomes a more complex mixed-integer programming or nonlinear programming problem. In some embodiments, the length of the discretized time slices can be adaptively adjusted according to the rate of load change, using shorter slices when changes are drastic. Optionally, a regulation cost term can be added to the objective function to form a multi-objective optimization, i.e., simultaneously minimizing the deviation and the total regulation cost. Optionally, the initial population generation of the iterative search algorithm can utilize the results of the load forecasting model to accelerate convergence.

[0042] In practical implementation, when using iterative search algorithms, such as genetic algorithms, each chromosome encodes a complete load allocation scheme, i.e., the power setpoints of all load resources across all time segments. The fitness function is based on the objective function. The algorithm calculates and penalizes solutions that violate constraints. Iteratively evolves the population through selection, crossover, and mutation operations, ultimately outputting the load allocation scheme corresponding to the chromosome with the highest fitness as the optimal control instruction. This optimal control instruction is formatted into a machine-readable instruction sequence and sent to the virtual power plant control platform for execution. It is understandable that the accuracy of the load forecasting model directly affects the quality of the optimal control instruction; an accurate forecast enables the optimization model to make decisions in a future scenario closer to reality, thereby reducing deviations during actual execution.

[0043] Example 5: In practical implementation, parsing the optimized control command is a crucial step before execution. The parsing process first decomposes the load adjustment time and deviation control requirements in the optimized control command. The optimized control command is a structured data object, typically containing information such as the command identifier, target load curve, execution time window, and allowable power deviation range. The decomposition operation extracts these key fields from the command. The load adjustment time specifies the exact start and end times of the load adjustment action, while the deviation control requirements quantify the maximum instantaneous and cumulative deviations between the actual load and the target load allowed during execution. The time limits of the load adjustment time and deviation control requirements are compared, and the minimum value is taken as the control cycle benchmark. For example, if the load adjustment time requires the adjustment to be completed within 10 minutes, while the deviation control requires the system to check for deviation exceeding the limit every 2 minutes, then the control cycle benchmark is set to 2 minutes. This means that the entire adjustment process will be divided into multiple control cycles with 2-minute intervals. At the end of each control cycle, effect evaluation and command fine-tuning are performed. The establishment of the control cycle benchmark provides a time scale basis for constructing a refined control action time sequence.

[0044] In practical implementation, a cost weight configuration is defined, including economic weight, stability weight, and reliability weight. The economic weight represents the emphasis on the economic costs incurred in the regulation process; the stability weight represents the emphasis on maintaining the stability of the system voltage and frequency; and the reliability weight represents the emphasis on the success rate of accurately and reliably executing load regulation commands. The specific values ​​of the cost weight configuration are manually set by the system operator based on the overall operating conditions of the current power grid and the priority of demand response commands, or automatically generated by the upper-level decision-making module. For example, in an emergency situation where the power grid frequency is close to the limit, the stability weight will be set to the highest value, and the economic weight will be reduced accordingly; while in conventional peak shaving and valley filling scenarios, the economic weight may dominate. The cost weight configuration is a three-dimensional vector, with values ​​typically between 0 and 1, and the sum of the three values ​​is 1 to ensure comparability across different dimensions. Based on the cost weight configuration, a comprehensive evaluation function is constructed. This function is used to quantitatively evaluate the comprehensive cost of a set of control parameters. Its output value is a scalar; the lower the value, the better the comprehensive performance of the set of control parameters. The construction process involves normalizing the quantitative indicators of the three dimensions of economy, stability, and reliability, and then weighting and combining them with their corresponding weights.

[0045] In practical implementation, control cases with similar scenarios are retrieved from historical operation records. These historical operation records are a database storing detailed records of various load regulation tasks previously performed by the virtual power plant. Each control case includes sample control parameters and sample execution costs. Sample control parameters are the set of control parameters actually used in historical operations, including specific values ​​such as load adjustment rate, adjustment duration, and stability maintenance interval. Load adjustment rate refers to the rate of load change per unit time; adjustment duration refers to the total time required from the start of regulation to reaching the target load; and stability maintenance interval refers to the shortest time required to maintain load stability after reaching the target load. Sample execution costs are the actual cost measures incurred after executing this set of sample control parameters, including energy consumption, equipment wear and tear, and grid deviation integral. Energy consumption refers to the total electrical energy consumed during regulation; equipment wear and tear is the estimated equipment lifespan depreciation based on the number of equipment switching operations and the magnitude of load changes; and grid deviation integral is the integral value of the deviation between the actual load curve and the target load curve, reflecting the tracking accuracy. The search for similar scenarios is based on the characteristics of the current optimized control instructions, such as matching similar target load changes, similar initial system states, and similar combinations of adjustable resources. The retrieved control cases constitute a case library for reference.

[0046] In practice, a comprehensive evaluation function is applied to assess control cases and select optimal control parameters. The evaluation process involves substituting the sample control parameters and sample execution costs of each retrieved historical control case into the comprehensive evaluation function to calculate a comprehensive score. (Comprehensive evaluation function) One specific form can be expressed as: in: This represents the overall evaluation value of the control case. The smaller the value, the lower the overall cost and the better the performance of the case. Represents the economic weight, which is specified by the cost weight configuration. The quantitative value representing the economic dimension is calculated based on the energy consumption and equipment wear and tear in the sample execution cost, and is usually normalized. This represents the stability weight, which is specified by the cost weight configuration. The quantitative value representing the stability dimension is calculated based on the grid deviation integral in the sample execution cost, reflecting the degree of impact on the grid, and is also normalized. Represents the reliability weight, which is specified by the cost weight configuration. The quantitative value representing the reliability dimension can be assigned based on historical records such as whether the goal was successfully achieved and whether there were any interruptions. The comprehensive evaluation value is calculated after all candidate cases are completed. Then, select The sample control parameters corresponding to the control case with the smallest value are taken as the optimal control parameters for the current scenario. These optimal control parameters embody control strategies that have proven effective under similar conditions in historical experience.

[0047] In practice, the control cycle benchmark is combined with the optimal control parameters to generate a control action time series. This time series is a detailed operation plan that specifies which load resources require which operations within each control cycle, from the start to the end of regulation. The combination process uses the control cycle benchmark as the time step, dividing the entire load adjustment time into several consecutive intervals. Within each interval, the target power value for the load resources is calculated according to the load adjustment rate defined in the optimal control parameters. Simultaneously, considering deviation control requirements, a power deviation warning line is set for that interval. The stability maintenance interval is also included in the sequence, manifested in the generation of a series of control commands to maintain the load level after the target load value is reached. The generated control action time series is typically in list or matrix format, with each row representing a control action and including information such as the timestamp of the action, the identifier of the controlled load resource, the target power value, and the allowable deviation range. This sequence is ultimately deployed to the virtual power plant control platform, where the platform's time scheduler executes it sequentially to complete the load regulation task.

[0048] See Figure 5 This graph, with control time on the horizontal axis and power on the vertical axis, presents the dynamic tracking relationship between target power and actual power during load regulation in a virtual power plant. Its core function is to verify the execution effect of the control action time series, demonstrating the adaptability of the adjustment rate in the control parameters. The graph intuitively reflects the technical value of using a comprehensive evaluation function to select optimal control parameters and combining the control cycle benchmark with the optimal parameters to generate a time series. The high degree of fit between actual power and target power verifies the accuracy of the optimized control commands and also demonstrates the responsiveness and reliability of the virtual power plant's load regulation.

[0049] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0050] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A virtual power plant load optimization control method based on demand response, characterized in that, The method includes: The data acquisition and control system automatically acquires the real-time operation dataset of the virtual power plant and the grid demand response instructions. The real-time operation dataset includes load power sequence and adjustable load resource information, while the grid demand response instructions include response time interval and load adjustment target value. Perform data cleaning and format standardization operations on the real-time running dataset to remove invalid data and unify data scale, and generate a preprocessed dataset. Mining load behavior characteristics from the preprocessed dataset, including load change rate and fluctuation cycle characteristics; By integrating load behavior characteristics and grid demand response commands, load resources are adapted, and a draft load control strategy is output. Configure the control node parameters of the virtual power plant according to the draft load control strategy, and collect the performance attribute data of the control nodes; Based on performance attribute data and draft load control strategies, simulation tools are used to calculate system operating constraints and define power variation boundaries and deviation tolerances. By using system operating constraints as optimization conditions, load allocation is solved to obtain optimized control commands; The control commands are analyzed and optimized, the control action time series is constructed, and the control action time series is deployed to the virtual power plant control platform to perform load regulation.

2. The demand-response-based virtual power plant load optimization control method as described in claim 1, characterized in that, Mining load behavior characteristics from the preprocessed dataset includes: A time-dimensional analysis was performed on the preprocessed dataset to extract the long-term trend component and short-term fluctuation component of the load power sequence. Statistical indicators for calculating short-term fluctuation components, including standard deviation and extreme value distribution; Identify the periodic patterns of the load power sequence and use spectral analysis to determine the dominant frequency component; By integrating statistical indicators of long-term trend components, short-term fluctuation components, and periodic patterns, a set of load behavior characteristics is formed.

3. The demand-response-based virtual power plant load optimization control method as described in claim 1, characterized in that, The draft output load control strategy includes load resource adaptation and includes: Establish a load resource rule base, which contains mapping relationships between various load types and regulation capabilities; The similarity between load behavior characteristics and load adjustment target values ​​is calculated to generate a matching score; Candidate load resources in the adjustable load resource information are filtered based on the matching score; Perform capacity verification on candidate load resources to ensure that the response time interval requirements are met; Combine candidate load resources and their regulation capabilities to generate a draft load control strategy.

4. The demand-response-based virtual power plant load optimization control method as described in claim 1, characterized in that, The system operational constraints are calculated using simulation tools, including: Load control node parameters and performance attribute data into the simulation environment; Set up simulation scenarios to simulate the system response under different load adjustment ranges; Collect simulation output data, including voltage deviation and frequency variation data; Analyze voltage deviation and frequency variation data to determine the critical conditions for stable system operation; The power variation boundary and deviation tolerance are derived based on the critical conditions.

5. The demand-response-based virtual power plant load optimization control method as described in claim 1, characterized in that, The load distribution solution is performed to obtain the optimal control commands, including: A load optimization model is constructed, with the objective function being to minimize the grid deviation. Inject power variation boundary and deviation tolerance as constraints into the load optimization model; The discretized response time interval is divided into multiple time segments; For each time segment, load adjustment amounts are allocated, and the feasible solution space is calculated based on adjustable load resource information. An iterative search algorithm is used to find the optimal solution in the feasible solution space and output optimized control instructions.

6. The demand-response-based virtual power plant load optimization control method as described in claim 5, characterized in that, Before building the load optimization model, a load forecasting model should be pre-built, including: Historical load records associated with load behavior characteristics and performance attribute data are extracted from the historical database. These historical load records include sample load command sequences, sample adjustment result sequences, deviation change sequences, and power change sequences. Historical load records are analyzed over time series, and the data are divided into multiple subsequences at fixed time intervals, with each subsequence serving as a training sample. For each training sample, the first and second derivatives of the power change sequence are calculated to extract the power change gradient features. At the same time, the mean and variance of the deviation change sequence are calculated to extract the deviation fluctuation features. By combining the power change gradient features and the deviation fluctuation features, a multi-dimensional feature vector is generated; A convolutional neural network model is trained using multidimensional feature vectors as input data. The convolutional neural network model contains convolutional layers, pooling layers, and fully connected layers. The convolutional layers are used to extract local temporal patterns, the pooling layers are used for dimensionality reduction, and the fully connected layers are used to output the predicted load value. The parameters of the convolutional neural network model are iteratively updated using the gradient descent algorithm to minimize the mean square error between the predicted load value and the actual load value. When the training error reaches a preset threshold, training stops, and the load prediction model is obtained.

7. The demand-response-based virtual power plant load optimization control method as described in claim 1, characterized in that, The parsing and optimization control instructions include: Decompose and optimize the load adjustment time and deviation control requirements in the control instructions; Compare the load adjustment time and the time limit of the deviation control requirement, and take the minimum value as the control cycle benchmark; Define the cost weight configuration, which includes economic weight, stability weight, and reliability weight; Construct a comprehensive evaluation function based on cost weighting; Retrieve control cases with similar scenarios from historical operation records. Each control case includes sample control parameters and sample execution cost. The sample control parameters include load adjustment rate, adjustment duration, and stability maintenance interval. The sample execution cost includes energy consumption, equipment loss degree, and grid deviation integral. The comprehensive evaluation function is used to evaluate the control case and select the optimal control parameters. By combining the control cycle benchmark with the optimal control parameters, a control action time series is generated.

8. The demand-response-based virtual power plant load optimization control method as described in claim 1, characterized in that, The data acquisition and control system automatically acquires real-time operational datasets of the virtual power plant and grid demand response commands, including: Deploy a distributed sensor network to continuously collect operational data from the load units; Receive demand response signals issued by the power grid center through the communication interface; Verify the integrity and consistency of operational data and demand response signals, and filter outliers; Aggregate operational data and demand response signals to form a standardized data flow.

9. The demand-response-based virtual power plant load optimization control method according to claim 3, characterized in that, The construction of the comprehensive evaluation function based on cost weight configuration is performed according to the following steps: Define evaluation dimensions, which include economic dimension, stability dimension and reliability dimension; Each evaluation dimension is assigned a weighting coefficient, which is dynamically adjusted according to the priority of the power grid demand response command. From an economic perspective, quantify the changes in energy costs associated with load adjustment; Regarding the stability dimension, the impact of load fluctuations on system frequency is quantified; For reliability, the expected success rate of load regulation is quantified; The quantized values ​​of different dimensions are normalized to eliminate the difference in dimensions. The normalized quantized values ​​are weighted and summed using weighting coefficients to generate a comprehensive evaluation function. The comprehensive evaluation function is calibrated using historical operational data to optimize the weight coefficient values.

10. The demand-response-based virtual power plant load optimization control method according to claim 4, characterized in that, The analysis of voltage deviation and frequency variation data determines the critical conditions for stable system operation, and is carried out according to the following steps: Extract the voltage deviation sequence and frequency variation sequence from the simulation output data; Set the voltage deviation threshold and frequency change threshold respectively; Traverse the voltage deviation sequence to identify extreme points that exceed the voltage deviation threshold and their duration; Traverse the frequency change sequence to identify extreme points that exceed the frequency change threshold and their duration; Calculate the system stability margin by associating the voltage extreme points and frequency extreme points at the same time section; Using the state where the system stability margin approaches zero as the critical condition, record the corresponding load adjustment range combination under this state; The boundary parameters for stable system operation are derived by combining the load adjustment range.

Images