A virtual power plant load forecasting and dynamic regulation optimization system and method

Abstract

This invention relates to the field of power plant data processing technology, and more particularly to a virtual power plant load forecasting and dynamic regulation optimization system and method, comprising a data acquisition module, a preprocessing module, a forecasting module, a regulation module, and a verification module. The data acquisition module acquires real-time operating data of distributed energy nodes and electricity market signals; the preprocessing module standardizes the data using quantum spatiotemporal alignment and anomaly reconstruction techniques, extracting strong correlation vectors between meteorological features and load; the forecasting module uses an adaptive noise complete set empirical mode decomposition algorithm to separate the load sequence trend term, periodic term, and residual components; and the regulation module constructs a multi-objective optimization model. This invention achieves efficient aggregation of distributed resources, high-precision load forecasting under meteorological change scenarios, and coordinated dynamic scheduling strategies across multiple markets, thereby improving the clean energy absorption capacity and market response efficiency of the virtual power plant.

Description

Technical Field

[0001] This invention relates to the field of power plant data processing technology, and in particular to a virtual power plant load forecasting and dynamic regulation optimization system and method. Background Technology

[0002] A virtual power plant is an intelligent energy dispatching platform built by aggregating heterogeneous resources such as distributed energy sources, energy storage systems, and flexible loads through advanced information and communication technologies and control systems. Its core relies on the Internet of Things (IoT) to achieve real-time monitoring of equipment status. Utilizing big data analytics, multi-source heterogeneous data fusion, and optimization algorithms, it completes supply and demand forecasting, dynamic optimization dispatching, and real-time response within the energy management system. It can participate in grid peak shaving and frequency regulation through demand-side response, and also provide ancillary services in the electricity market, achieving collaborative and intelligent operation of distributed resources and significantly improving grid flexibility and renewable energy absorption capacity.

[0003] In the field of virtual power plant technology, existing systems often face problems of low resource aggregation efficiency and insufficient regulation capacity when integrating distributed energy resources due to the diversity of resource types, geographical dispersion, and dynamic complexity of the power system. Existing methods have limited accuracy in load forecasting, making it difficult to accurately match grid demand response requirements, resulting in insufficient clean energy absorption capacity and a lack of dynamic optimization mechanisms adaptable to multiple market scenarios. Furthermore, existing technologies have limitations in optimizing the operational strategies of virtual power plants participating in electricity market transactions, failing to effectively coordinate real-time interaction between the source and load sides and market risk response, thus restricting the flexibility and economy of virtual power plants in electricity ancillary services and the spot market. These shortcomings urgently require systemic technological improvements to achieve efficient synergy between load forecasting and dynamic regulation, thereby enhancing the overall operational efficiency and market adaptability of virtual power plants. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a virtual power plant load forecasting and dynamic regulation optimization system and method. This system solves the problems of insufficient aggregation efficiency caused by the diversity and decentralization of distributed resources in existing virtual power plant systems, the inability of existing load forecasting models to meet the requirements of high-precision grid response, and the lack of adaptability of dynamic regulation mechanisms to various market scenarios, which restrict the consumption of clean energy and the efficiency of market transactions.

[0005] To solve the above-mentioned technical problems, the specific technical solution of the present invention is as follows:

[0006] In a first aspect, the present invention provides a virtual power plant load forecasting and dynamic regulation optimization system, comprising:

[0007] The data acquisition module is used to acquire real-time operating data and electricity market signals from distributed energy nodes, and transmit the real-time operating data to the preprocessing module;

[0008] The preprocessing module is configured to preprocess real-time running data and generate standardized time-series data to be output to the prediction module;

[0009] The forecasting module is configured to decompose standardized time series data, separate the trend term, periodic term and residual component of the load sequence, input the trend term, periodic term and residual component into the meteorological and load coupling model, the meteorological and load coupling model integrates meteorological characteristics and decomposed component sequences, generates load forecast curves for future preset periods and transmits them to the regulation module.

[0010] The regulation module is configured to receive load forecast curves and real-time electricity market price signals, establish a multi-objective optimization model with the goals of maximizing renewable energy consumption rate, minimizing operating costs, and achieving frequency regulation response qualification rate, use a genetic algorithm to screen the optimal scheduling scheme, and generate source-load coordinated scheduling instructions including power allocation instructions to be sent to the execution terminal.

[0011] The verification module is configured to receive the deviation data between the actual scheduling data and the load forecast curve fed back by the execution terminal, identify the contribution of meteorological sensitivity features and equipment response lag features in the deviation data, mark features whose contribution exceeds a preset threshold, and generate feature weight update instructions to feed back to the preprocessing module; at the same time, it performs multi-scenario simulation of the source-load coordinated scheduling instructions through a preset quantum digital twin model, and generates prediction model parameter correction instructions and objective function constraint condition update instructions based on the equipment output deviation and market revenue deviation in the simulation results. The prediction model parameter correction instructions are used to update the optimization parameters of the meteorological and load coupling model, and the constraint condition update instructions are used to correct the decision variable boundaries of the multi-objective optimization model.

[0012] Furthermore, in the virtual power plant load forecasting and dynamic regulation optimization system of the present invention, the preprocessing module is also used for:

[0013] The system receives power fluctuation data from energy nodes and raw meteorological monitoring data transmitted from the data acquisition module, performs outlier cleaning and missing value filling on the power data, and forms an initial processed data stream.

[0014] The meteorological data and power data in the initial data stream are timestamped and then normalized using a sliding window to generate standardized time-series data.

[0015] Normalized meteorological feature data is labeled with meteorological sensitivity tags and formed into two independent channels with standardized power time series data. The meteorological feature data is directly transmitted to the prediction module, while the power time series data is transmitted in parallel to the prediction module and the cloud storage module.

[0016] Furthermore, in the virtual power plant load forecasting and dynamic regulation optimization system of the present invention, the forecasting module is also used for:

[0017] When receiving the decomposed trend term, periodic term and residual component sequence output by the preprocessing module, the component sequence feature vectors associated with the current time period stored in the historical database are called synchronously.

[0018] Based on the pattern matching results of the decomposed component sequences and historical feature vectors, an evaluation result reflecting the stability of component fluctuations is generated.

[0019] The confidence interval range of the load forecast curve is dynamically adjusted based on the stability assessment results, and the forecast curve with confidence level markers is synchronously output to the adjustment module and the verification module.

[0020] Furthermore, in the virtual power plant load forecasting and dynamic regulation optimization system of the present invention, the regulation module is also used for:

[0021] While receiving real-time electricity price signals from the electricity market, the system simultaneously acquires grid reserve capacity declaration data and real-time status data of demand response contracts to form a scheduling decision input dataset.

[0022] Based on the real-time monitoring results of the intensity of electricity price signal fluctuations, the time granularity control mode of the scheduling strategy is dynamically switched, wherein when a sudden change in electricity price is detected, the minute-level scheduling instruction generation mode is activated.

[0023] The scheduling scheme instruction set generated based on the multi-objective optimization model is submitted to the verification module for integrity verification. After the verification is passed, a request for issuing instructions with timestamps is sent to the execution terminal.

[0024] Furthermore, in the virtual power plant load forecasting and dynamic regulation optimization system of the present invention, the verification module is also used for:

[0025] Receive device response delay data returned by the execution terminal, extract the corresponding original meteorological data sampling values ​​within the device response delay data, and establish a basis for correlation analysis between delay characteristics and meteorological conditions;

[0026] The device response delay data and the theoretical response curve output by the prediction model are convolved in the time domain to generate a characteristic fingerprint vector that represents the dynamic response of the device.

[0027] When the deviation between the feature fingerprint vector and the baseline vector in the historical feature library exceeds a preset threshold, a feature weight update instruction set including time-sensitive factors is generated, and a dynamic weight matrix update request is sent to the preprocessing module to optimize subsequent data processing.

[0028] Furthermore, in the virtual power plant load forecasting and dynamic regulation optimization system of the present invention, the execution terminal is also used for:

[0029] When receiving the source-load coordinated scheduling command issued by the regulation module, the real-time state of charge data of the local energy storage device is obtained simultaneously to form a set of command execution constraints.

[0030] Based on the timing requirements of the power allocation command and the state of charge data, the timing phase of the command execution is dynamically adjusted, and equipment protection constraints are injected to generate a safe execution command sequence.

[0031] The adjusted sequence of actual executed instructions is bound to the device operating status data to generate a feedback data packet with a checksum, which is then sent back to the verification module for closed-loop verification via the communication interface.

[0032] Secondly, this invention provides a method for virtual power plant load forecasting and dynamic regulation optimization, applied to the aforementioned virtual power plant load forecasting and dynamic regulation optimization system, comprising the following steps:

[0033] S101: Obtain real-time operating data and electricity market signals from distributed energy nodes, and transmit the real-time operating data to S102;

[0034] S102, preprocess the real-time running data, generate standardized time-series data and output it to S103;

[0035] S103: The standardized time series data is decomposed using a preset decomposition algorithm to obtain the trend term, periodic term and residual component of the load sequence. The residual component and the light intensity feature are input into the quantum long short-term memory network. The dynamic change feature of wind speed is fused through the quantum self-attention mechanism to generate the load prediction curve and output to S104.

[0036] S104: Receive load forecast curve and real-time electricity market price signal, use genetic algorithm to solve multi-objective Pareto solution set of absorption rate, operating cost and frequency regulation response qualification rate, generate source-load coordinated scheduling instruction including power allocation instruction through preset sorting algorithm, send it to execution terminal and trigger S105.

[0037] S105 receives the deviation data between the actual scheduling data and the load forecast curve fed back by the execution terminal, identifies the contribution of meteorological sensitivity features and equipment response lag features in the deviation data, marks features whose contribution exceeds a preset threshold, and generates feature weight update instructions to feed back to the preprocessing module; at the same time, it performs multi-scenario simulation of the source-load coordinated scheduling instructions through a preset quantum digital twin model, and generates prediction model parameter correction instructions and objective function constraint condition update instructions based on the equipment output deviation and market revenue deviation in the simulation results. The parameter correction instructions are used to update the particle swarm optimization parameters in the meteorological and load coupling model, and the constraint condition update instructions are used to correct the decision variable boundaries of the multi-objective optimization model.

[0038] Beneficial effects of this invention;

[0039] This invention effectively improves the efficiency of distributed resource aggregation and the dynamic response capability of the system through a modular architecture and a multi-source data collaborative processing mechanism. The quantum spatiotemporal alignment and anomaly reconstruction technology in the preprocessing module solves the problem of time-series data deviation caused by sudden environmental changes or equipment heterogeneity, enhancing the spatiotemporal correlation accuracy between meteorological characteristics and load fluctuations. The adaptive decomposition algorithm and quantum long short-term memory network in the prediction module, combined with a meteorological coupling model and dynamic adjustment of confidence intervals, significantly improve the adaptability of load forecasting to complex meteorological conditions and reduce the risk of grid response deviation. The multi-objective optimization model and minute-level dispatch instruction generation mechanism in the regulation module integrate real-time signals from the electricity market and equipment state constraints to achieve coordinated optimization of renewable energy absorption rate, operating costs, and frequency regulation response. The verification module, through digital twin simulation and characteristic fingerprint vector analysis, constructs a closed-loop feedback link for iterative prediction model parameters and dynamic update of feature weights, improving the execution reliability of dispatch strategies and the efficiency of clean energy absorption in multi-market scenarios. Each module is based on quantum algorithms and cross-regional federated learning protocols, forming a full-process optimization system from data acquisition and strategy generation to execution verification, systematically breaking through the technical bottlenecks of traditional virtual power plants in terms of resource aggregation, prediction accuracy and market adaptability. Attached Figure Description

[0040] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on the drawings without creative effort.

[0041] Figure 1 The flowchart illustrates a virtual power plant load prediction and dynamic regulation optimization method provided in this embodiment of the invention.

[0042] Figure 2 This is a system architecture diagram for virtual power plant load forecasting and dynamic regulation optimization provided in an embodiment of the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention. The technical solutions provided by various embodiments of this invention will be described in detail below with reference to the accompanying drawings. To better understand the objectives of this invention, it will be described in further detail below.

[0044] Firstly, please refer to Figure 2This invention provides a virtual power plant load forecasting and dynamic regulation optimization system, comprising:

[0045] The data acquisition module is used to acquire real-time operating data and electricity market signals from distributed energy nodes, and transmit the real-time operating data to the preprocessing module;

[0046] The preprocessing module is configured to preprocess real-time running data and generate standardized time-series data to be output to the prediction module;

[0047] The forecasting module is configured to decompose standardized time series data, separate the trend term, periodic term and residual component of the load sequence, input the trend term, periodic term and residual component into the meteorological and load coupling model, the meteorological and load coupling model integrates meteorological characteristics and decomposed component sequences, generates load forecast curves for future preset periods and transmits them to the regulation module.

[0048] The regulation module is configured to receive load forecast curves and real-time electricity market price signals, establish a multi-objective optimization model with the goals of maximizing renewable energy consumption rate, minimizing operating costs, and achieving frequency regulation response qualification rate, use a genetic algorithm to screen the optimal scheduling scheme, and generate source-load coordinated scheduling instructions including power allocation instructions to be sent to the execution terminal.

[0049] The verification module is configured to receive the deviation data between the actual scheduling data and the load forecast curve fed back by the execution terminal, identify the contribution of meteorological sensitivity features and equipment response lag features in the deviation data, mark features whose contribution exceeds a preset threshold, and generate feature weight update instructions to feed back to the preprocessing module; at the same time, it performs multi-scenario simulation of the source-load coordinated scheduling instructions through a preset quantum digital twin model, and generates prediction model parameter correction instructions and objective function constraint condition update instructions based on the equipment output deviation and market revenue deviation in the simulation results. The prediction model parameter correction instructions are used to update the optimization parameters of the meteorological and load coupling model, and the constraint condition update instructions are used to correct the decision variable boundaries of the multi-objective optimization model.

[0050] This invention provides a virtual power plant load forecasting and dynamic regulation optimization system and method, aiming to improve the aggregation efficiency of distributed energy resources and the dynamic regulation capability of the power grid, and solve problems such as insufficient load forecasting accuracy and rigid regulation mechanisms in traditional technologies. The system achieves closed-loop optimization of data acquisition, processing, forecasting, scheduling, and verification through modular design, specifically including the following core modules:

[0051] The data acquisition module is responsible for acquiring real-time operational data (such as photovoltaic output and energy storage status) and electricity market signals (such as electricity price and frequency regulation demand) from distributed energy nodes, providing raw input for subsequent processing.

[0052] The preprocessing module standardizes the collected data, including timestamp alignment, outlier removal, and missing data imputation. It also extracts the correlation between meteorological parameters and load characteristics to generate enhanced feature vectors to improve prediction accuracy.

[0053] The forecasting module uses a decomposition algorithm to split the load sequence into trend components, periodic components, and residual components. It then combines meteorological data to construct a coupled model, generates load forecast curves for future periods, and dynamically adjusts model parameters through an adaptive mechanism to cope with sudden meteorological changes.

[0054] Based on load forecast curves and real-time electricity price signals, the regulation module establishes a multi-objective optimization model with renewable energy absorption rate, operating cost and frequency regulation response qualification rate as optimization objectives. It uses intelligent algorithms to generate the optimal scheduling scheme and coordinate the power allocation of resources such as energy storage and distributed power sources.

[0055] The verification module receives the deviation information between the actual scheduling data and the prediction results, locates the source of the deviation through contribution analysis, and uses digital twin technology to simulate the scheduling effect under multiple scenarios, dynamically corrects the prediction model parameters and optimizes the model constraints, forming a closed-loop feedback mechanism.

[0056] Furthermore, the system enhances regional resource coordination capabilities, real-time communication, energy storage equipment lifespan, and demand-side response flexibility through extended modules such as cross-regional collaboration, communication compensation, lifespan management, and user interaction. Based on quantum algorithms and dynamic feedback mechanisms, each module enables the virtual power plant to achieve accurate prediction, efficient scheduling, and risk management in complex environments, significantly improving clean energy consumption rates and market responsiveness.

[0057] Specifically, in the virtual power plant load forecasting and dynamic regulation optimization system of the present invention, the preprocessing module is further used for:

[0058] The system receives power fluctuation data from energy nodes and raw meteorological monitoring data transmitted from the data acquisition module, performs outlier cleaning and missing value filling on the power data, and forms an initial processed data stream.

[0059] The meteorological data and power data in the initial data stream are timestamped and then normalized using a sliding window to generate standardized time-series data.

[0060] Normalized meteorological feature data is labeled with meteorological sensitivity tags and formed into two independent channels with standardized power time series data. The meteorological feature data is directly transmitted to the prediction module, while the power time series data is transmitted in parallel to the prediction module and the cloud storage module.

[0061] The preprocessing module undertakes the core functions of data cleaning and standardization in the system. The specific process is as follows:

[0062] The preprocessing module first receives power fluctuation data from energy nodes and raw meteorological monitoring data from the data acquisition module. It removes invalid data caused by equipment failure or sudden environmental changes through anomaly detection and cleaning mechanisms, and intelligently fills in missing data to form a continuous and complete initial processing data stream.

[0063] Subsequently, the module aligns the timestamps of meteorological and power data to eliminate data timing misalignment caused by differences in the clocks of the acquisition equipment or transmission delays. It also uses a sliding window normalization method to standardize the time series data, eliminating the impact of dimensional differences on subsequent analysis and generating a standardized time series dataset in a unified format.

[0064] After processing, the module labels the normalized meteorological feature data with meteorological sensitivity tags and separates it from the standardized power time series data into two independent channels. The meteorological feature data is directly transmitted to the forecasting module for load coupling analysis, while the power time series data is synchronously transmitted to the forecasting module for decomposition modeling and backed up to the cloud storage module for historical data retrieval and model training optimization. This design ensures both data processing efficiency and provides multi-dimensional data support for the system.

[0065] Specifically, in the virtual power plant load forecasting and dynamic regulation optimization system of the present invention, the forecasting module is further used for:

[0066] When receiving the decomposed trend term, periodic term and residual component sequence output by the preprocessing module, the component sequence feature vectors associated with the current time period stored in the historical database are called synchronously.

[0067] Based on the pattern matching results of the decomposed component sequences and historical feature vectors, an evaluation result reflecting the stability of component fluctuations is generated.

[0068] The confidence interval range of the load forecast curve is dynamically adjusted based on the stability assessment results, and the forecast curve with confidence level markers is synchronously output to the adjustment module and the verification module.

[0069] The forecasting module incorporates historical data comparison and dynamic evaluation mechanisms during the load forecasting process to further optimize the reliability of the forecast results. The specific process is as follows:

[0070] When the prediction module receives the trend term, periodic term and residual component sequence after decomposition by the preprocessing module, it simultaneously associates the component feature vectors of similar time periods in the historical database, analyzes the difference in fluctuation patterns between the current component sequence and the historical data through pattern matching, and evaluates the stability level of each component.

[0071] Based on the stability assessment results, the module dynamically adjusts the confidence interval range of the load forecast curve: for components with high volatility (such as residual terms), the confidence interval is appropriately widened to reflect forecast uncertainty; for components with high stability (such as periodic terms), the confidence interval is narrowed to improve forecast accuracy. The final generated load forecast curve includes confidence level markers, which intuitively represent the forecast reliability at different time periods.

[0072] The adjusted prediction curve is synchronously transmitted to the adjustment module and the verification module. The adjustment module optimizes the fault-tolerant design of the scheduling strategy based on the confidence information, while the verification module uses the confidence level to locate the high-incidence periods of prediction deviation, forming a closed loop of linkage optimization between prediction accuracy and scheduling strategy.

[0073] Specifically, in the virtual power plant load forecasting and dynamic regulation optimization system of the present invention, the regulation module is further used for:

[0074] While receiving real-time electricity price signals from the electricity market, the system simultaneously acquires grid reserve capacity declaration data and real-time status data of demand response contracts to form a scheduling decision input dataset.

[0075] Based on the real-time monitoring results of the intensity of electricity price signal fluctuations, the time granularity control mode of the scheduling strategy is dynamically switched, wherein when a sudden change in electricity price is detected, the minute-level scheduling instruction generation mode is activated.

[0076] The scheduling scheme instruction set generated based on the multi-objective optimization model is submitted to the verification module for integrity verification. After the verification is passed, a request for issuing instructions with timestamps is sent to the execution terminal.

[0077] The adjustment module, based on multi-source data fusion and dynamic strategy adjustment mechanisms, enables the accurate generation and execution of scheduling schemes. The specific process is as follows:

[0078] While receiving real-time electricity price signals from the electricity market, the regulation module simultaneously integrates dynamic information such as grid reserve capacity application data and demand response contract execution status to form a multi-dimensional scheduling decision input dataset, providing real-time data support for the optimization model.

[0079] The adjustment module dynamically adjusts the time granularity control mode of the scheduling strategy by monitoring the fluctuation intensity of the electricity price signal in real time: when a sudden change in electricity price or abnormal market fluctuation is detected, the minute-level high-frequency scheduling instruction generation mode is automatically activated to quickly respond to market changes; during the electricity price stable phase, the conventional time granularity strategy is adopted to balance scheduling efficiency and computing resource consumption.

[0080] The generated scheduling scheme instruction set needs to be submitted to the verification module for logical integrity verification, including instruction compliance, timing consistency, and resource matching checks. After successful verification, the adjustment module sends a timestamped instruction issuance request to the execution terminal to ensure the timing accuracy and execution traceability of the scheduling instructions, thereby improving the system's response speed and reliability to dynamic changes in the electricity market.

[0081] Specifically, in the virtual power plant load forecasting and dynamic regulation optimization system of the present invention, the verification module is further used for:

[0082] Receive device response delay data returned by the execution terminal, extract the corresponding original meteorological data sampling values ​​within the device response delay data, and establish a basis for correlation analysis between delay characteristics and meteorological conditions;

[0083] The device response delay data and the theoretical response curve output by the prediction model are convolved in the time domain to generate a characteristic fingerprint vector that represents the dynamic response of the device.

[0084] When the deviation between the feature fingerprint vector and the baseline vector in the historical feature library exceeds a preset threshold, a feature weight update instruction set including time-sensitive factors is generated, and a dynamic weight matrix update request is sent to the preprocessing module to optimize subsequent data processing.

[0085] The verification module improves the adaptability of the system's data processing and prediction models through device response delay analysis and dynamic feature optimization mechanisms. The specific process is as follows:

[0086] After receiving the device response delay data from the execution terminal, the verification module synchronously extracts the original meteorological data sampling values ​​(such as temperature, wind speed, etc.) within the corresponding time period, constructs a dataset for correlation analysis between delay characteristics and meteorological conditions, and identifies the potential impact patterns of environmental factors on device response lag.

[0087] The verification module performs temporal convolution calculations between the device response delay data and the theoretical response curve output by the prediction model to generate a characteristic fingerprint vector that characterizes the device's dynamic response properties. This vector is used to quantify the degree of deviation between the actual operating state of the device and the theoretical expectation of the prediction model, forming a traceable basis for evaluating response characteristics.

[0088] When the deviation between the characteristic fingerprint vector and the baseline vector stored in the historical feature database exceeds a preset threshold, the verification module generates a feature weight update instruction set including time-sensitive factors and sends a dynamic weight matrix update request to the preprocessing module. This process optimizes the fusion weight allocation of meteorological features and equipment response features in subsequent data processing, enhancing the system's ability to adapt to complex meteorological conditions and dynamic equipment responses.

[0089] Specifically, in the virtual power plant load forecasting and dynamic regulation optimization system of the present invention, the execution terminal is further used for:

[0090] When receiving the source-load coordinated scheduling command issued by the regulation module, the real-time state of charge data of the local energy storage device is obtained simultaneously to form a set of command execution constraints.

[0091] Based on the timing requirements of the power allocation command and the state of charge data, the timing phase of the command execution is dynamically adjusted, and equipment protection constraints are injected to generate a safe execution command sequence.

[0092] The adjusted sequence of actual executed instructions is bound to the device operating status data to generate a feedback data packet with a checksum, which is then sent back to the verification module for closed-loop verification via the communication interface.

[0093] The execution terminal implements security control and data closed-loop management during the execution and feedback of scheduling instructions. The specific process is as follows:

[0094] When the terminal receives the source-load coordinated scheduling command issued by the regulation module, it simultaneously obtains the real-time charge status data of the local energy storage device (such as remaining capacity, charge and discharge rate, etc.), and combines it with the device operating parameters to form a set of dynamic constraints for command execution, ensuring the compatibility of scheduling commands with the actual operating conditions of the device.

[0095] Based on the timing requirements of power allocation commands and real-time state of charge data, the terminal dynamically adjusts the time phase offset of command execution to eliminate execution timing conflicts caused by equipment response lag or sudden state changes. Simultaneously, it injects equipment protection constraints such as overcharge / discharge protection thresholds and temperature limits to generate an execution command sequence that meets safety requirements, balancing scheduling objectives with equipment operational risks.

[0096] The adjusted sequence of actual executed instructions is bound to equipment operating status data (such as real-time power output and voltage fluctuation range) to generate a feedback data packet carrying a unique checksum. The data packet is transmitted back to the verification module through the communication interface to complete the matching verification between the actual execution result and the scheduling instruction, thus constructing a closed-loop control link from instruction generation and execution to feedback verification, and enhancing the reliability and traceability of system operation.

[0097] Secondly, please refer to Figure 1 This invention provides a method for virtual power plant load forecasting and dynamic regulation optimization, applied to a virtual power plant load forecasting and dynamic regulation optimization system, comprising the following steps:

[0098] S101: Obtain real-time operating data and electricity market signals from distributed energy nodes, and transmit the real-time operating data to S102;

[0099] S102, preprocess the real-time running data, generate standardized time-series data and output it to S103;

[0100] S103: The standardized time series data is decomposed using a preset decomposition algorithm to obtain the trend term, periodic term and residual component of the load sequence. The residual component and the light intensity feature are input into the quantum long short-term memory network. The dynamic change feature of wind speed is fused through the quantum self-attention mechanism to generate the load prediction curve and output to S104.

[0101] S104: Receive load forecast curve and real-time electricity market price signal, use genetic algorithm to solve multi-objective Pareto solution set of absorption rate, operating cost and frequency regulation response qualification rate, generate source-load coordinated scheduling instruction including power allocation instruction through preset sorting algorithm, send it to execution terminal and trigger S105.

[0102] S105 receives the deviation data between the actual scheduling data and the load forecast curve fed back by the execution terminal, identifies the contribution of meteorological sensitivity features and equipment response lag features in the deviation data, marks features whose contribution exceeds a preset threshold, and generates feature weight update instructions to feed back to the preprocessing module; at the same time, it performs multi-scenario simulation of the source-load coordinated scheduling instructions through a preset quantum digital twin model, and generates prediction model parameter correction instructions and objective function constraint condition update instructions based on the equipment output deviation and market revenue deviation in the simulation results. The parameter correction instructions are used to update the particle swarm optimization parameters in the meteorological and load coupling model, and the constraint condition update instructions are used to correct the decision variable boundaries of the multi-objective optimization model.

[0103] The technical solution of this invention addresses the shortcomings of existing virtual power plant systems by achieving systematic improvements through the following technical approach:

[0104] This invention employs a modular architecture design, integrating data acquisition, preprocessing, prediction, and regulation modules to construct a standardized processing flow for multi-source heterogeneous data. The preprocessing module utilizes quantum spatiotemporal alignment and anomaly reconstruction techniques to eliminate time-series data deviations caused by sudden environmental changes or communication delays. It also filters for strong correlations between meteorological conditions and load based on quantized mutual information entropy, enhancing the spatiotemporal consistency of distributed resource output characteristics. The cross-regional collaboration module aggregates historical data from multiple regions through a federated learning protocol to jointly train prediction models, optimizing decomposition algorithm parameters and addressing the insufficient aggregation efficiency caused by resource dispersion.

[0105] The prediction module employs an adaptive noise-complete ensemble empirical mode decomposition algorithm to separate the trend term, periodic term, and residual components of the load sequence. It then dynamically fuses environmental variables such as wind speed and light intensity using a meteorological and load coupling model. A quantum long short-term memory network embeds a quantum self-attention mechanism to dynamically adjust the weights of the residual components based on meteorological abrupt changes, generating a load prediction curve with confidence intervals. The verification module simulates prediction deviations across multiple scenarios using a quantum digital twin model. It iteratively updates model parameters based on error source analysis results, forming a self-optimizing closed loop for prediction accuracy, meeting the high-precision response requirements of the power grid.

[0106] The regulation module establishes a multi-objective optimization model centered on renewable energy absorption rate, operating costs, and frequency regulation response qualification rate. It employs an improved genetic algorithm to solve for the Pareto front solution set and generates minute-level dispatch instructions by combining real-time electricity price signals and market reserve capacity data. The market response unit decomposes the energy storage capacity allocation strategy using a quantum game theory model, dynamically balancing spot arbitrage and frequency regulation reserve demand. The verification module simultaneously performs integrity checks on dispatch instructions and analyzes equipment response delays. Based on the deviation of characteristic fingerprint vectors, it dynamically updates the feature weight matrix, optimizes strategy generation constraints, and improves the system's dynamic regulation capability and clean energy absorption efficiency in various electricity market scenarios.

[0107] The technical features of this invention are explained below:

[0108] Quantum spatiotemporal alignment algorithm (preprocessing module): solves the problem of data timing misalignment caused by geographical differences and device clock drift in distributed nodes.

[0109] Implementation logic: A time synchronization mechanism is constructed based on the principle of quantum entanglement to entangle and associate the clock signals of meteorological sensors and energy node metering equipment.

[0110] When the timestamp deviation between meteorological data and power data exceeds a threshold of 200 milliseconds, the quantum state synchronization protocol is activated to align the data timeline with nanosecond-level precision.

[0111] The output is aligned temporal data stream, supporting the spatiotemporal consistency of subsequent feature extraction.

[0112] Adaptive Noise Complete Set Empirical Mode Decomposition (CEEMDAN, Prediction Module): solves the problem of separating trend terms, periodic terms and random disturbance terms in load sequences.

[0113] Implementation logic: Adaptive white noise is injected into the original load sequence, and modal aliasing is eliminated through multiple iterative decompositions. The sample entropy criterion is used to automatically identify the number of decomposition layers and separate: trend term (low-frequency component, representing long-term load changes), periodic term (mid-frequency component, reflecting intraday / weekly patterns), and residual term (high-frequency random fluctuations). The fluctuation stability assessment results of the output component sequence are then dynamically adjusted (±2%~5%).

[0114] Quantum Long Short-Term Memory Network (QLSTM) and Self-Attention Mechanism (Prediction Module): Improving the Robustness of Load Prediction under Sudden Weather Scenarios.

[0115] Implementation logic:

[0116] QLSTM structure: The neurons of the traditional LSTM are replaced with qubit gates, enhancing the ability to express nonlinear features. In residual component processing, the quantum state superposition mechanism captures transient features of sudden changes in light intensity (such as cloud cover).

[0117] Quantum self-attention mechanism: Calculates the similarity weight (range 0-1) between dynamic wind speed change characteristics and historical patterns. When a sudden change in wind speed is detected (change >5m / s within 3 seconds), its weight is automatically increased to above 0.7.

[0118] Multi-objective optimization model and genetic algorithm (adjustment module): coordinating conflicting objectives such as renewable energy consumption rate, operating cost, and frequency regulation response qualification rate.

[0119] Implementation logic:

[0120] Objective function design:

[0121] Objective 1: Maximize the absorption rate (actual renewable energy output / predicted renewable energy generation).

[0122] Objective 2: Minimize operating costs (electricity purchase cost + equipment depreciation cost);

[0123] Objective 3: Maximize the frequency regulation qualification rate (|actual output - command requirement| ≤ 2%).

[0124] Improved genetic algorithm:

[0125] Pareto solution sets are generated using the NSGA-III framework;

[0126] Introduce an electricity price volatility factor (compress the optimization generation to 50 generations when volatility > 10%).

[0127] Output minute-level scheduling commands (e.g., energy storage charging and discharging power accuracy of ±10MW).

[0128] Quantum digital twin model and characteristic fingerprint analysis (verification module): Constructing a closed-loop correction mechanism for device response deviation.

[0129] Implementation logic:

[0130] Twin model simulation:

[0131] A quantum state superposition model of 3,000 equipment operation scenarios was constructed based on historical data;

[0132] The equipment output deviation is simulated under boundary conditions such as sudden changes in market electricity prices and extreme weather.

[0133] Feature fingerprint vector generation:

[0134] Extract time-frequency domain features (such as delay phase and amplitude attenuation) from device response delay data.

[0135] A 128-dimensional fingerprint vector is generated by temporal convolution.

[0136] Deviation analysis: When the cosine similarity between the fingerprint vector and the historical baseline is less than 0.8, feature weight updates are triggered.

[0137] Federated Learning Protocol (Cross-Regional Collaboration): Enables collaborative optimization of models across multiple regions while ensuring data privacy.

[0138] Implementation logic:

[0139] The preprocessing module uploads the encrypted feature vector (AES-256 encryption) to the regional server;

[0140] The server aggregates gradient information from 100 nodes and updates the QLSTM network weights.

[0141] Each node downloads the updated model parameters and executes the inference task locally.

[0142] Example 1: Load prediction for a sudden weather change scenario:

[0143] This system was deployed in a virtual power plant in region A. The data acquisition module acquired power data from 20 distributed photovoltaic nodes and monitoring data from 7 meteorological stations at a sampling frequency of 5 seconds. The preprocessing module detected a lack of instantaneous irradiance due to severe convective weather and used a quantum spatiotemporal alignment algorithm based on Gaussian process regression to complete 10% data imputation and timestamp calibration within 150 milliseconds. The normalized irradiance abrupt change characteristics (fluctuations >800W / m² within 10 minutes) were marked as high-sensitivity tags and transmitted to the prediction module. The prediction module called the CEEMDAN algorithm to decompose the load sequence. When the fluctuation amplitude of the residual component exceeded 15% of the historical baseline value, the meteorological abrupt change response unit in the quantum long short-term memory network was activated, dynamically increasing the attention weight of the wind speed feature to 0.75. A load forecast curve for the next 4 hours was generated. Actual operation verified that the prediction error rate was reduced in a scenario with a sudden drop in irradiance of 40%.

[0144] Example 2: Multi-market collaborative scheduling optimization

[0145] During the participation of the virtual power plant in Area B in the day-ahead energy market and the real-time frequency regulation market, the regulation module synchronously receives load forecast curves (updated every 15 minutes), spot electricity prices (every 2 seconds), and AGC frequency regulation commands. When the price volatility exceeds the 12% threshold at 14:30, a minute-level dispatch mode is activated. A multi-objective optimization model, constrained by a grid connection rate ≥85%, a frequency regulation qualification rate ≥95%, and cost minimization, uses an improved genetic algorithm to generate a Pareto solution set within 45 seconds. The optimal solution allocates the energy storage system to participate in spot arbitrage with a power discharge of 15MW during the peak electricity price period (14:35-14:40), while reserving 40% capacity to respond to frequency regulation demand. Actual operation shows that the grid connection rate increases and frequency regulation revenue increases during this period.

[0146] Example 3: Equipment Response Lag Compensation

[0147] In the wind farm cluster access system of Region C, the execution terminal detected an 8-second delay in charging and discharging of a 2MW energy storage device due to temperature protection. The verification module extracted the wind speed abrupt change characteristics during the delay period (increasing from 6m / s to 14m / s within 3 seconds), calculating that the contribution of the device's response lag characteristics reached 38%. Using a quantum digital twin model, 2000 sets of extreme operating conditions were simulated, generating characteristic fingerprint vectors and comparing them with historical benchmarks, showing a deviation of 25%. Based on this, a feature weight update command was issued: the preprocessing module increased the feature weight of the wind speed change rate from 0.3 to 0.45; the adjustment module tightened the SOC operating boundary of this type of energy storage device to 25%-80%. In subsequent scheduling, the response delay rate of similar devices decreased.

[0148] Example 4: Cross-regional federated learning optimization:

[0149] A virtual power plant alliance covering multiple provinces and cities implemented cross-regional collaboration. The preprocessing module uploaded encrypted meteorological-sensitive feature vectors to the regional collaboration server, where 52 nodes jointly conducted federated learning training. After 48 hours of joint training: the accuracy of the QLSTM network in the prediction module in identifying humidity characteristics during the plum rain season improved; the regulation module reconstructed and optimized the model based on regional electricity price fluctuation patterns, shortening the dispatch instruction generation time. Under continuous rainy weather, the system load forecast RMSE remained stable within the expected range, and the day-ahead market clearing deviation decreased.

[0150] Example 5: Verification of High-Proportion Consumption of New Energy:

[0151] In the D City Renewable Energy Demonstration Zone, the system connects to a virtual power plant where wind / solar power account for 83% of the installed capacity. The verification module monitors the fluctuation of the curtailment rate in real time. When the solar curtailment rate suddenly increases to 9.7% during a certain period: the contribution of meteorological sensitive features (high-frequency fluctuations in irradiance caused by cloud movement) reaches 52%, the feature is marked and the prediction model is retrained; the quantum digital twin simulation shows that the energy storage frequency regulation reserve constraint needs to be relaxed, and the lower limit of capacity allocation should be adjusted from 20% to 15%. Under similar meteorological conditions the day after the strategy update, the curtailment rate decreases and the intraday frequency regulation response qualification rate improves.

[0152] The implementation of this invention is based on a modular system architecture, where modules work together to achieve efficient aggregation of distributed resources, high-precision load forecasting, and dynamic scheduling optimization across multiple markets. The system is deployed on a cloud computing platform and connects to distributed energy nodes (photovoltaic power plants, energy storage systems, flexible loads, etc.) via industrial IoT protocols, collecting power fluctuation data and meteorological monitoring data in real time. The data acquisition module obtains real-time operating parameters of the nodes at an adjustable sampling frequency of 1 second to 5 minutes, simultaneously subscribing to spot electricity price signals and frequency regulation ancillary service demand signals from the electricity market trading platform, forming a multi-source heterogeneous data input stream.

[0153] The preprocessing module employs a quantized timestamp alignment algorithm to eliminate clock drift errors between meteorological sensors and power metering equipment. For the power data stream, outliers are identified based on the Laida criterion and imputed using Gaussian process regression; meteorological data undergoes noise reduction through sliding window mean filtering. Normalization uses the z-score standardization method to generate two independent data channels: feature data labeled with meteorological sensitivity tags (temperature, irradiance, wind speed) are directly input into the prediction module; standardized power time-series data are synchronously transmitted to the prediction module and the cloud-based historical database.

[0154] The prediction module invokes the Adaptive Noise Complete Ensemble Empirical Mode Decomposition (CEEMDAN) algorithm to decompose the power time series data into trend, periodic, and residual component sequences. The residual components are processed using a Quantum Long Short-Term Memory (QLSTM) network, and a quantum self-attention mechanism is introduced to dynamically weight and fuse wind speed abrupt change features and light intensity features. The model performs a rolling prediction every 15 minutes, generating a load curve for the next 4 hours. Simultaneously, the confidence interval (85%–95%) is dynamically adjusted based on historical component feature vector matching results, outputting prediction results with probability distribution characteristics.

[0155] After receiving the load forecast curve, the regulation module combines real-time electricity price signals (updated every 2 seconds) and grid reserve capacity declaration data to construct a multi-objective optimization model. The objective functions are set as follows: maximizing the renewable energy absorption rate (weight coefficient 0.4), minimizing operating costs (weight coefficient 0.3), and maximizing the frequency regulation response qualification rate (weight coefficient 0.3). An improved genetic algorithm (population size 200, crossover probability 0.85) is used to solve the Pareto solution set, and the optimal scheduling scheme is generated through TOPSIS sorting. When the electricity price volatility exceeds the 10% threshold, the minute-level scheduling mode is activated, generating a source-load coordination instruction set that includes energy storage charging and discharging power and photovoltaic output adjustment instructions.

[0156] The verification module receives actual output data from the execution terminal via the OPC UA protocol and calculates the root mean square error (RMSE) of the predicted curve. The Shapley value decomposition algorithm is used to identify the sources of deviation: when the contribution of meteorologically sensitive features (such as sudden rain or snow) is >40%, the feature is marked and a weight update instruction is generated; when the contribution of equipment response lag features (such as energy storage charging and discharging delay) is >30%, the quantum digital twin model is triggered to perform multi-scenario simulations. The twin model constructs 3000 simulation scenarios based on historical operating data, outputting an analysis report on equipment output deviation and market revenue deviation. Based on this, the particle swarm optimization parameters of the meteorological load coupling model are updated (the learning factor is adjusted from 0.5 to 0.7), and the decision variable boundaries of the multi-objective optimization model are corrected (e.g., the energy storage SOC constraint range is tightened from 20%~90% to 30%~85%).

[0157] The execution terminal deploys an edge computing unit locally. After receiving scheduling instructions, it collects energy storage SOC data in real time (accuracy ±0.5%). Based on the instruction timing requirements and SOC status, a phase delay of 1-5 minutes is dynamically inserted to match the device response characteristics, and constraints such as over-temperature protection (derating above 45℃) and over-voltage protection (shutdown above 1.1 pu) are superimposed to generate a safety instruction sequence. Finally, the timestamped execution data is bound to the device status, encrypted with AES-256, and fed back to the verification module, forming a closed-loop optimization link.

[0158] The system performs federated learning training every 24 hours: the preprocessing module uploads encrypted feature vectors to the regional collaborative server, and 100 distributed nodes jointly update the QLSTM network weights; the adjustment module reconstructs and optimizes the model constraint matrix based on historical scheduling data to improve adaptability to multiple market scenarios. Through the above implementation methods, the system achieves a reduction in load forecasting error rate, an increase in renewable energy consumption rate, and an improvement in frequency regulation response qualification rate under meteorological change scenarios.

Claims

1. A virtual power plant load forecasting and dynamic regulation optimization system, characterized in that, include: The data acquisition module is used to acquire real-time operating data and electricity market signals from distributed energy nodes, and transmit the real-time operating data to the preprocessing module; The preprocessing module is configured to preprocess real-time running data and generate standardized time-series data to be output to the prediction module; The forecasting module is configured to decompose standardized time-series data, separate the trend term, periodic term, and residual component of the load sequence, input the trend term, periodic term, and residual component into the meteorological and load coupling model, process the residual component through a quantum long short-term memory network, and introduce a quantum self-attention mechanism to dynamically weight and fuse wind speed change characteristics and light intensity characteristics. The meteorological and load coupling model fuses meteorological characteristics and decomposed component sequences to generate a load forecast curve for a future preset period and transmit it to the regulation module. The regulation module is configured to receive load forecast curves and real-time electricity market price signals, establish a multi-objective optimization model with the goals of maximizing renewable energy consumption rate, minimizing operating costs, and achieving frequency regulation response qualification rate, use a genetic algorithm to screen the optimal scheduling scheme, and generate source-load coordinated scheduling instructions including power allocation instructions to be sent to the execution terminal. The verification module is configured to receive the deviation data between the actual scheduling data and the load forecast curve fed back by the execution terminal, identify the contribution of meteorological sensitivity features and equipment response lag features in the deviation data, mark features whose contribution exceeds a preset threshold, generate feature weight update instructions and feed them back to the preprocessing module, perform multi-scenario simulation of the source-load coordinated scheduling instructions through a preset quantum digital twin model, and generate prediction model parameter correction instructions and objective function constraint condition update instructions based on the equipment output deviation and market revenue deviation in the simulation results. The prediction model parameter correction instructions are used to update the optimization parameters of the meteorological and load coupling model, and the constraint condition update instructions are used to correct the decision variable boundaries of the multi-objective optimization model.

2. The virtual power plant load forecasting and dynamic regulation optimization system according to claim 1, characterized in that, The preprocessing module is also used for: The system receives power fluctuation data from energy nodes and raw meteorological monitoring data transmitted from the data acquisition module, performs outlier cleaning and missing value filling on the power data, and forms an initial processed data stream. The meteorological data and power data in the initial data stream are timestamped and then normalized using a sliding window to generate standardized time-series data. Normalized meteorological feature data is labeled with meteorological sensitivity tags and formed into two independent channels with standardized power time series data. The meteorological feature data is directly transmitted to the prediction module, while the power time series data is transmitted in parallel to the prediction module and the cloud storage module.

3. The virtual power plant load forecasting and dynamic regulation optimization system according to claim 2, characterized in that, The prediction module is also used for: When receiving the decomposed trend term, periodic term and residual component sequence output by the preprocessing module, the component sequence feature vectors associated with the current time period stored in the historical database are called synchronously. Based on the pattern matching results of the decomposed component sequences and historical feature vectors, an evaluation result reflecting the stability of component fluctuations is generated. The confidence interval range of the load forecast curve is dynamically adjusted based on the stability assessment results, and the forecast curve with confidence level markers is synchronously output to the adjustment module and the verification module.

4. The virtual power plant load forecasting and dynamic regulation optimization system according to claim 3, characterized in that, The adjustment module is also used for: While receiving real-time electricity price signals from the electricity market, the system simultaneously acquires grid reserve capacity declaration data and real-time status data of demand response contracts to form a scheduling decision input dataset. Based on the real-time monitoring results of the intensity of electricity price signal fluctuations, the time granularity control mode of the scheduling strategy is dynamically switched, wherein when a sudden change in electricity price is detected, the minute-level scheduling instruction generation mode is activated. The scheduling scheme instruction set generated based on the multi-objective optimization model is submitted to the verification module for integrity verification. After the verification is passed, a request for issuing instructions with timestamps is sent to the execution terminal.

5. The virtual power plant load forecasting and dynamic regulation optimization system according to claim 4, characterized in that, The verification module is also used for: Receive device response delay data returned by the execution terminal, extract the corresponding original meteorological data sampling values ​​within the device response delay data, and establish a basis for correlation analysis between delay characteristics and meteorological conditions; The device response delay data and the theoretical response curve output by the prediction model are convolved in the time domain to generate a characteristic fingerprint vector that represents the dynamic response of the device. When the deviation between the feature fingerprint vector and the baseline vector in the historical feature library exceeds a preset threshold, a feature weight update instruction set including time-sensitive factors is generated, and a dynamic weight matrix update request is sent to the preprocessing module to optimize subsequent data processing.

6. The virtual power plant load forecasting and dynamic regulation optimization system according to claim 5, characterized in that, The execution terminal is further used for: When receiving the source-load coordinated scheduling command issued by the regulation module, the real-time state of charge data of the local energy storage device is obtained simultaneously to form a set of command execution constraints. Based on the timing requirements of the power allocation command and the state of charge data, the timing phase of the command execution is dynamically adjusted, and equipment protection constraints are injected to generate a safe execution command sequence. The adjusted sequence of actual executed instructions is bound to the device operating status data to generate a feedback data packet with a checksum, which is then sent back to the verification module for closed-loop verification via the communication interface.

7. A method for virtual power plant load forecasting and dynamic regulation optimization, applied to the virtual power plant load forecasting and dynamic regulation optimization system as described in any one of claims 1 to 6, characterized in that, Includes the following steps: S101: Obtain real-time operating data and electricity market signals from distributed energy nodes, and transmit the real-time operating data to S102; S102, preprocess the real-time running data to generate standardized time-series data and output it to S103; S103: The standardized time series data is decomposed using a preset decomposition algorithm to obtain the trend term, periodic term and residual component of the load sequence. The residual component is processed by a quantum long short-term memory network, and a quantum self-attention mechanism is introduced to dynamically weight and fuse the wind speed change feature and the light intensity feature to generate a load prediction curve and output it to S104. S104: Receive load forecast curve and real-time electricity market price signal, use genetic algorithm to solve multi-objective Pareto solution set of absorption rate, operating cost and frequency regulation response qualification rate, generate source-load coordinated scheduling instruction including power allocation instruction through preset sorting algorithm, send it to execution terminal and trigger S105. S105 receives the deviation data between the actual scheduling data and the load forecast curve fed back by the execution terminal, identifies the contribution of meteorological sensitivity features and equipment response lag features in the deviation data, marks features whose contribution exceeds a preset threshold and generates feature weight update instructions to feed back to the preprocessing module, performs multi-scenario simulation of the source-load coordinated scheduling instructions through a preset quantum digital twin model, and generates prediction model parameter correction instructions and objective function constraint condition update instructions based on the equipment output deviation and market revenue deviation in the simulation results. The parameter correction instructions are used to update the particle swarm optimization parameters in the meteorological and load coupling model, and the constraint condition update instructions are used to correct the decision variable boundaries of the multi-objective optimization model.

Images