Virtual power plant energy scheduling method and system based on artificial intelligence

Abstract

The invention belongs to the technical field of energy scheduling, and particularly discloses a virtual power plant energy scheduling method and system based on artificial intelligence, and the method comprises the steps: collecting the multi-modal features of a virtual power plant, constructing an elastic correction equation, combining with a historical electricity price load data correction coefficient, and outputting a power prediction value and an elastic correction value of the virtual power plant. The method comprises the following steps: acquiring real-time characteristic data and physical constraints of a virtual power plant, performing three-party dynamic gaming in combination with a power predicted value and an elastic correction of the virtual power plant, acquiring real-time state data of the virtual power plant, outputting an optimization instruction of the virtual power plant, and analyzing quantum annealing fault tolerance based on the optimization instruction and real-time abrupt change characteristic data of the virtual power plant. And determining a dynamic adjustment rule, and outputting a final execution signal of the virtual power plant. According to the invention, by comprehensively integrating the multi-modal features and real-time data of the virtual power plant, high-precision power prediction, intelligent resource scheduling and dynamic strategy adjustment are realized, and the accuracy, flexibility and stability of energy management are significantly improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of energy scheduling, and specifically relates to a virtual power plant energy scheduling method and system based on artificial intelligence. Background Art

[0002] Renewable energy generation is characterized by intermittent and fluctuating characteristics. For example, wind power is affected by wind speed, and photovoltaic power is affected by the duration and intensity of sunlight. Large-scale integration of renewable energy will pose challenges to the stable operation of the power system, requiring more intelligent scheduling methods to balance supply and demand. Virtual power plants can aggregate multiple energy resources, such as distributed generation, energy storage, and controllable loads, to achieve synergy between source, grid, load, and storage.

[0003] However, existing energy management systems often employ localized and decentralized strategies, lacking cross-regional and cross-industry coordination and integration capabilities. This has led to problems such as imbalanced energy supply and demand, inefficient transmission, and insufficient utilization of energy storage resources. With the large-scale integration of renewable energy, effectively managing the intermittent and distributed nature of these energy sources to ensure a stable and secure energy supply has become an unresolved issue.

[0004] Artificial intelligence technologies such as machine learning, deep learning, and reinforcement learning are developing rapidly, with significant achievements in areas such as image recognition and natural language processing. These technologies can be considered for use in virtual power plant data processing, model training, and strategy optimization, providing technical means for energy scheduling. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides an artificial intelligence-based virtual power plant energy scheduling method and system. By integrating advanced artificial intelligence technologies such as long short-term memory network (LSTM), elastic correction mechanism, three-party dynamic game, manifold optimization and quantum annealing fault tolerance, it comprehensively integrates the multimodal characteristics and real-time data of the virtual power plant, realizes high-precision power prediction, intelligent resource scheduling and dynamic strategy adjustment, significantly improves the accuracy, flexibility and stability of energy management, and optimizes energy utilization efficiency.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: The energy dispatching method of a virtual power plant based on artificial intelligence includes the following steps: Collect multimodal features of the virtual power plant, perform standard LSTM operations, and determine the correction coefficient of historical electricity price and load data; Construct an elasticity correction equation, combine it with the correction coefficient of historical electricity price and load data, and output the virtual power plant power forecast value and elasticity correction amount; Acquire the real-time characteristic data and physical constraints of the virtual power plant, combine the virtual power plant power forecast value and elasticity correction value to conduct a three-party dynamic game, and obtain the equilibrium strategy set and risk gradient matrix of the virtual power plant; Collect real-time status data of the virtual power plant, perform manifold optimization based on the equilibrium strategy set and risk gradient matrix of the virtual power plant, and output optimization instructions for the virtual power plant; Collect the real-time mutation feature data of the virtual power plant, analyze the quantum annealing fault tolerance based on the optimization instructions and real-time mutation feature data of the virtual power plant, determine the dynamic adjustment rules, and output the final execution signal of the virtual power plant.

[0007] Preferably, the multimodal features of the virtual power plant are collected and a standard LSTM operation is performed, and the specific process is as follows: Get the hidden state of the LSTM at time t-1 stored in the database and the cell state at time t-1 ; Collect multimodal characteristics of virtual power plants, including photovoltaic power of virtual power plants , Virtual power plant fan speed , virtual power plant irradiance , virtual power plant wind turbine power ; Perform standard LSTM operations: Forget Gate: ; Where, is the forget gate output at time t, is the sigmoid activation function, is the forget gate weight matrix, is the input at the current time t, is the forget gate bias term; Input Gate: ; Where, is the input gate output at time t, is the input gate weight matrix, is the input gate bias term; Candidate cell states: ; Where, is the state output of the candidate cell at time t, is the candidate cell state weight matrix, is the candidate cell state bias; Cell status update: ; Where, is the cell state output at time t; Output Gate: ; Where, is the output gate output at time t, is the output gate weight matrix, is the output gate bias term; Basic hidden state: ; Where, is the basic hidden state at time t; Physical gradient term injection: ; Where, is the physical gradient term, is the physical gradient term scaling factor stored in the database, Calibration factor for the IV curve of monocrystalline silicon modules stored in the database, is the fan characteristic curve fitting factor stored in the database; Final status update: ; ; Where, is the rate of change of the hidden layer state h over time, is the hidden state of LSTM at time t; Obtain the hidden state of LSTM stored in the database - the historical electricity price load data correction coefficient mapping set, and determine the matching historical electricity price load data correction coefficient based on the hidden state of LSTM at the current time t .

[0008] Preferably, the elasticity correction equation is constructed, combined with the correction coefficient of historical electricity price and load data, to output the virtual power plant power forecast value and elasticity correction amount. The specific process is as follows: Collect the power sequence of the virtual power plant in the past 24 hours and real-time electricity prices for virtual power plants ; LSTM basic prediction: ; Where, The power value of the virtual power plant at time t obtained by LSTM-based prediction; Calculation of elastic correction term: ; Where, is the elastic correction, is the demand price elasticity coefficient stored in the database, is the rate of change of the real-time electricity price of the virtual power plant over time; Final prediction output: ; Where, is the predicted power value of the virtual power plant at time t.

[0009] Preferably, the real-time characteristic data and physical constraints of the virtual power plant are obtained by: Obtain real-time characteristic data of the virtual power plant, including the electricity price of the i-th node of the virtual power plant , Virtual Power Plant Energy Storage SOC Status , virtual power plant grid frequency deviation , i is the virtual power plant node number; Capture the physical constraints of the virtual power plant, including: Obtain the virtual power plant irradiance-photovoltaic maximum output mapping set stored in the database, and determine the matching photovoltaic maximum output based on the current virtual power plant irradiance ; Virtual power plant wind turbine ramp rate constraints ; in, is the wind turbine power of the virtual power plant, is the rated power of the virtual power plant wind turbine, is the rate of change of wind turbine power in the virtual power plant over time; The physical constraints of the virtual power plant are used as constraints to obtain the equilibrium strategy set of the virtual power plant; The electricity price and maximum photovoltaic output of each node of the virtual power plant are stored as specified tags, and the specified tag-photovoltaic operation and maintenance cost mapping set stored in the database is obtained. Based on the current specified tag, the matching photovoltaic operation and maintenance cost is determined. .

[0010] Preferably, the three-party dynamic game is conducted by combining the virtual power plant power forecast value and the elasticity correction amount to obtain the equilibrium strategy set and risk gradient matrix of the virtual power plant. The specific process is: Three-party dynamic game modeling: The participants are: PV operators, energy storage systems, and power grids; PV operator's corresponding strategy variable x: adjust the power generation ; Energy storage system corresponding strategy variable y: control charging and discharging power ; Grid corresponding strategy variable z: adjust the grid interaction power ; Strategy Space: ; Where, is the predicted power value of the virtual power plant at time t; ; Where, is the maximum discharge power stored in the database, The maximum charging power stored in the database; ; Where, is the minimum grid interaction power stored in the database, is the maximum grid interaction power stored in the database; Potential energy function : ; Where, is the potential energy function corresponding to the three strategy variables x, y, and z, is the photovoltaic operation and maintenance cost coefficient stored in the database, is the energy storage aging penalty coefficient stored in the database, is the first frequency sensitivity coefficient stored in the database, is the second frequency sensitivity coefficient stored in the database, and e is a natural constant; The ultimate goal is to minimize the total potential energy; Solve the Nash equilibrium and iteratively update the strategy until convergence: ; In the formula, k is the number of iterations, is the value of variable x after the k+1th iteration, is the value of variable y after the k+1th iteration, is the value of variable z after the k+1th iteration, is the value of variable x at the kth iteration, is the value of variable y at the kth iteration, is the value of variable z at the kth iteration, is the step length; After convergence, output the equilibrium strategy set of the virtual power plant ; is the target power generation, is the target charge and discharge power, is the target grid interaction power; Risk Gradient Matrix for: .

[0011] Preferably, the real-time status data of the virtual power plant is collected, and the equilibrium strategy set and risk gradient matrix of the virtual power plant are combined to perform manifold optimization, and the optimization instructions of the virtual power plant are output. The specific process is: Collect real-time status data of the virtual power plant, including the voltage of the i-th node of the virtual power plant , virtual power plant line load rate ; Riemannian metric construction: ; In the formula, x is the strategy variable corresponding to the photovoltaic operator, y is the strategy variable corresponding to the energy storage system, and z is the strategy variable corresponding to the power grid. is the photovoltaic cost weight, is the energy storage life weight stored in the database, is the grid stability weight stored in the database, is the line load penalty term stored in the database, s is the arc length parameter on the Riemannian manifold; The geodesic equation is numerically solved using the fourth-order Runge-Kutta method: ; Where, is the risk gradient matrix, 、 、 is the tensor index, Corresponding power generation, Corresponding charge and discharge power, Corresponding to the grid interaction power, is the Christoffel symbol, t is the time index; Dynamic weight adjustment: ; Where, is the photovoltaic cost weight at time t, is the initial photovoltaic cost weight stored in the database, e is a natural constant, is the frequency deviation of the virtual power plant grid at time t; Optimization instructions for virtual power plants generate: ; Where, To optimize power generation, To optimize the charge and discharge power, To optimize grid interaction power.

[0012] Preferably, the real-time mutation characteristic data of the virtual power plant is collected, and the specific process is as follows: Collect real-time mutation feature data of virtual power plants, including fault signs of virtual power plant equipment , Virtual Power Plant Wind Speed Sudden Change Signal , virtual power plant frequency secondary deviation ; Where n is the number of virtual power plant devices.

[0013] Preferably, the optimization instructions based on the virtual power plant and the real-time mutation characteristic data are used to analyze the quantum annealing fault tolerance, and the specific process is as follows: Construct the Hamiltonian H: ; Where, is the parameter of the interaction strength between device p and device q in the system, is the component of the Pauli matrix of device p in the B direction, is the component of the Pauli matrix of device q in the B direction, is the transverse field intensity at time t, is the component of the Pauli matrix of device p in the A direction, is the fault penalty coefficient stored in the database, is the fault flag of the pth device, p and q are the device numbers, and t is the time index; Tunneling probability calculate: ; Where, is the reduced Planck constant, is the transverse field intensity, is the energy difference; Obtain the tunneling probability-time constant mapping set stored in the database, and determine the matching time constant based on the tunneling probability .

[0014] Preferably, the dynamic adjustment rules are determined and the final execution signal of the virtual power plant is outputted. The specific process is: Dynamic adjustment rules: Photovoltaic off-grid: ; Where, is the final charge and discharge power, To optimize the charge and discharge power, is the power variation stored in the database, e is a natural constant, and t is a time index; Frequency exceeds the limit: ; Where, is the power adjustment amount, is the scale factor stored in the database, is the integral coefficient stored in the database; Output the final execution signal of the virtual power plant: ; ; Where, To optimize power generation, is the final power generation; ; Where, To optimize grid interaction power, is the final grid interaction power.

[0015] The artificial intelligence-based virtual power plant energy dispatching system is used to implement the artificial intelligence-based virtual power plant energy dispatching method described above, including: Standard LSTM operation module, used to collect multimodal features of virtual power plants and perform standard LSTM operations; The elasticity correction equation construction module is used to construct the elasticity correction equation, combine the correction coefficient of historical electricity price and load data, and output the virtual power plant power forecast value and elasticity correction amount; The equilibrium strategy set acquisition module is used to obtain the real-time characteristic data and physical constraints of the virtual power plant, and conduct a three-party dynamic game based on the virtual power plant power forecast value and elasticity correction value to obtain the equilibrium strategy set and risk gradient matrix of the virtual power plant; The optimization instruction analysis module is used to collect real-time status data of the virtual power plant, perform manifold optimization based on the equilibrium strategy set and risk gradient matrix of the virtual power plant, and output the optimization instructions of the virtual power plant; The final execution signal output module is used to collect the real-time mutation feature data of the virtual power plant, analyze the quantum annealing fault tolerance based on the optimization instructions and real-time mutation feature data of the virtual power plant, determine the dynamic adjustment rules, and output the final execution signal of the virtual power plant.

[0016] The present invention has the following beneficial effects: This method collects multimodal features and performs standard LSTM operations. LSTM (Long Short-Term Memory) networks excel at processing time series data and can effectively capture the complex, time-dependent dynamic characteristics of distributed energy generation and load changes in virtual power plants. For example, these characteristics include fluctuations in photovoltaic output affected by sunlight intensity and duration, and the cyclical nature of user electricity usage habits. Combining these multimodal features enables more comprehensive and accurate power forecasting, providing a reliable basis for subsequent scheduling. A flexible correction equation is constructed to output power predictions and flexible corrections. Taking into account the uncertainties of renewable energy generation and load, the flexible correction dynamically adjusts the predicted values based on real-time data and model analysis, reducing prediction errors and ensuring that scheduling plans are more aligned with actual operating conditions.

[0017] The present invention combines multiple aspects of information to conduct a three-party dynamic game. Virtual power plants involve multiple stakeholders, including the power generation side, the grid side, and the user side, to achieve a reasonable allocation of resources among different stakeholders and improve overall operational efficiency. At the same time, the risk gradient matrix helps to assess and control scheduling risks. Manifold optimization is performed based on real-time status data and the previously obtained strategy set. Manifold learning can mine the inherent low-dimensional structure of the data to find a more optimal scheduling trajectory in the virtual power plant state space, making the optimization instructions more precise, improving energy utilization efficiency, and reducing unnecessary energy loss.

[0018] This invention combines optimization instructions with real-time mutation signature data to analyze quantum annealing fault tolerance and determine dynamic adjustment rules. This allows for rapid response and adjustment of scheduling strategies in the face of sudden changes in renewable energy generation, such as extreme weather events or sudden failures. Quantum annealing fault tolerance analysis ensures stable system operation under complex circumstances, while dynamic adjustment rules provide scheduling flexibility and adaptability, ensuring reliable power supply for the virtual power plant. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 Schematic diagram of the process of the present invention; Figure 2 Detailed flow chart of the method of the present invention; Figure 3 Schematic diagram of module connection of the system of the present invention. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the present invention will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present invention.

[0021] During calculation, each formula in this embodiment may be dimensionless as needed to simplify the calculation.

[0022] Example 1: Figure 1 、 Figure 2 As shown, the energy dispatching method of a virtual power plant based on artificial intelligence includes the following steps: Collect multimodal features of the virtual power plant, perform standard LSTM operations, and determine the correction coefficient of historical electricity price and load data; Construct an elasticity correction equation, combine it with the correction coefficient of historical electricity price and load data, and output the virtual power plant power forecast value and elasticity correction amount; Acquire the real-time characteristic data and physical constraints of the virtual power plant, combine the virtual power plant power forecast value and elasticity correction value to conduct a three-party dynamic game, and obtain the equilibrium strategy set and risk gradient matrix of the virtual power plant; Collect real-time status data of the virtual power plant, perform manifold optimization based on the equilibrium strategy set and risk gradient matrix of the virtual power plant, and output optimization instructions for the virtual power plant; Collect the real-time mutation feature data of the virtual power plant, analyze the quantum annealing fault tolerance based on the optimization instructions and real-time mutation feature data of the virtual power plant, determine the dynamic adjustment rules, and output the final execution signal of the virtual power plant.

[0023] Collect multimodal features of the virtual power plant and perform standard LSTM operations. The specific process is as follows: Get the hidden state of the LSTM at time t-1 stored in the database and the cell state at time t-1 ; Collect multimodal characteristics of virtual power plants, including photovoltaic power of virtual power plants , Virtual power plant fan speed , virtual power plant irradiance , virtual power plant wind turbine power ; Perform standard LSTM operations: Forget Gate: ; Where, is the forget gate output at time t, is the sigmoid activation function, is the forget gate weight matrix, is the input at the current time t, is the forget gate bias term; Input Gate: ; Where, is the input gate output at time t, is the input gate weight matrix, is the input gate bias term; Candidate cell states: ; Where, is the state output of the candidate cell at time t, is the candidate cell state weight matrix, is the candidate cell state bias; Cell status update: ; Where, is the cell state output at time t; Output Gate: ; Where, is the output gate output at time t, is the output gate weight matrix, is the output gate bias term; Basic hidden state: ; Where, is the basic hidden state at time t; Physical gradient term injection: ; Where, is the physical gradient term, is the physical gradient term scaling factor stored in the database, Calibration factor for the IV curve of monocrystalline silicon modules stored in the database, is the fan characteristic curve fitting factor stored in the database; Final status update: ; ; Where, is the rate of change of the hidden layer state h over time, is the hidden state of LSTM at time t; Obtain the hidden state of LSTM stored in the database - the historical electricity price load data correction coefficient mapping set, and determine the matching historical electricity price load data correction coefficient based on the hidden state of LSTM at the current time t .

[0024] LSTMs possess memory capabilities and can effectively process time series data through structures such as a forget gate, input gate, and output gate. The forget gate determines which information from the previous cell state is discarded, the input gate controls the inclusion of current input information, and the output gate determines the output content. In virtual power plants, they can effectively capture the dynamic changes in energy data over time, such as the time-dependent fluctuations in photovoltaic power and wind turbine power. They are better than traditional neural networks at handling long-term and short-term dependencies, improving forecasting accuracy.

[0025] Collect multimodal features of the virtual power plant, including photovoltaic power, wind turbine speed, irradiance, and wind turbine power. LSTM can comprehensively process these diverse and interrelated features to fully reflect the virtual power plant's operating status and explore potential connections between the data. For example, analyzing the mapping between irradiance and photovoltaic power allows for more accurate prediction of generated power.

[0026] By injecting physical gradient terms and combining the physical characteristics related to photovoltaic power and wind turbine power (determined by the IV curve calibration factor of the monocrystalline silicon module, the wind turbine characteristic curve fitting factor, etc.), the model is not only based on data learning, but also incorporates physical knowledge and actual operating laws, enhancing the model's understanding and simulation capabilities of the complex physical processes of the virtual power plant, reducing prediction bias, and improving model accuracy.

[0027] Construct an elasticity correction equation, combine it with the correction coefficient of historical electricity price and load data, and output the virtual power plant power forecast value and elasticity correction amount. The specific process is as follows: Collect the power sequence of the virtual power plant in the past 24 hours and real-time electricity prices for virtual power plants ; LSTM basic prediction: ; Where, The power value of the virtual power plant at time t obtained by LSTM-based prediction; Calculation of elastic correction term: ; Where, is the elastic correction, is the demand price elasticity coefficient stored in the database, is the rate of change of the real-time electricity price of the virtual power plant over time; Final prediction output: ; Where, is the predicted power value of the virtual power plant at time t.

[0028] The virtual power plant's power series over the past 24 hours is collected for basic LSTM forecasting. LSTM excels at processing time series and can identify patterns, periodicity, and trends in power data over time. For example, it can identify patterns in daily load curves and fluctuations in renewable energy generation. This provides a solid foundation for power forecasting and more accurately captures dynamic power changes than simpler methods.

[0029] Real-time electricity prices and their rate of change over time are incorporated into the elasticity correction calculation. Electricity prices reflect the relationship between electricity supply and demand. Real-time price changes can guide users to adjust their electricity usage, thereby affecting the output of the virtual power plant. By incorporating dynamic changes in electricity prices into the forecast, the model can capture the responsive relationship between electricity price and power, making the forecast more accurate to market conditions.

[0030] The LSTM basic prediction results are adjusted using elastic correction terms. Considering that virtual power plant operations are affected by a variety of uncertainties, elastic corrections can make real-time corrections to the basic predictions based on dynamic changes in electricity prices. This overcomes the limitations of historical power forecasts alone, reduces prediction errors, and brings the final prediction output closer to the actual power value, improving prediction accuracy.

[0031] Obtain real-time characteristic data and physical constraints of the virtual power plant. The specific process is as follows: Obtain real-time characteristic data of the virtual power plant, including the electricity price of the i-th node of the virtual power plant , Virtual Power Plant Energy Storage SOC Status , virtual power plant grid frequency deviation , i is the virtual power plant node number; Capture the physical constraints of the virtual power plant, including: Obtain the virtual power plant irradiance-photovoltaic maximum output mapping set stored in the database, and determine the matching photovoltaic maximum output based on the current virtual power plant irradiance ; Virtual power plant wind turbine ramp rate constraints ; in, is the wind turbine power of the virtual power plant, is the rated power of the virtual power plant wind turbine, is the rate of change of wind turbine power in the virtual power plant over time; The physical constraints of the virtual power plant are used as constraints to obtain the equilibrium strategy set of the virtual power plant; The electricity price and maximum photovoltaic output of each node of the virtual power plant are stored as specified tags, and the specified tag-photovoltaic operation and maintenance cost mapping set stored in the database is obtained. Based on the current specified tag, the matching photovoltaic operation and maintenance cost is determined. .

[0032] Combining the virtual power plant power forecast value and elasticity correction value to conduct a three-party dynamic game, the equilibrium strategy set and risk gradient matrix of the virtual power plant are obtained. The specific process is as follows: Three-party dynamic game modeling: The participants are: PV operators, energy storage systems, and power grids; PV operator's corresponding strategy variable x: adjust the power generation ; Energy storage system corresponding strategy variable y: control charging and discharging power ; Grid corresponding strategy variable z: adjust the grid interaction power ; Strategy Space: ; Where, is the predicted power value of the virtual power plant at time t; ; Where, is the maximum discharge power stored in the database, The maximum charging power stored in the database; ; Where, is the minimum grid interaction power stored in the database, is the maximum grid interaction power stored in the database; Potential energy function : ; Where, is the potential energy function corresponding to the three strategy variables x, y, and z, is the photovoltaic operation and maintenance cost coefficient stored in the database, is the energy storage aging penalty coefficient stored in the database, is the first frequency sensitivity coefficient stored in the database, is the second frequency sensitivity coefficient stored in the database, and e is a natural constant; The ultimate goal is to minimize the total potential energy; Solve the Nash equilibrium and iteratively update the strategy until convergence: ; In the formula, k is the number of iterations, is the value of variable x after the k+1th iteration, is the value of variable y after the k+1th iteration, is the value of variable z after the k+1th iteration, is the value of variable x at the kth iteration, is the value of variable y at the kth iteration, is the value of variable z at the kth iteration, is the step length; After convergence, output the equilibrium strategy set of the virtual power plant ; is the target power generation, is the target charge and discharge power, is the target grid interaction power; Risk Gradient Matrix for: .

[0033] Acquiring real-time characteristic data such as virtual power plant node electricity prices, energy storage SOC status, and grid frequency deviation covers electricity market price signals, energy storage equipment status, and key grid operation indicators. It can fully reflect the operating environment and status of virtual power plants and provide rich information for formulating reasonable strategies.

[0034] Taking into account physical constraints such as the maximum photovoltaic output and wind turbine ramp rate, ensure that the scheduling strategy is formulated within the actual operating capacity of the virtual power plant equipment, avoid equipment damage or inability to execute due to excessive scheduling, and ensure safe and stable operation of the system.

[0035] This dynamic game involves PV operators, energy storage systems, and the power grid. Different stakeholders have different interests and regulatory approaches. This game can coordinate their actions, leading to more rational allocation of resources such as PV generation, energy storage charging and discharging, and grid interaction power, thereby improving the overall operational efficiency and economic benefits of the virtual power plant.

[0036] The potential energy function comprehensively reflects factors such as photovoltaic operation and maintenance costs, energy storage aging penalties, and frequency deviation impacts, quantifying costs and risks of different natures into a unified indicator. This facilitates the evaluation and comparison of the comprehensive costs of the system under different strategies and provides a clear goal for strategy optimization.

[0037] By adopting an iterative update strategy until convergence to solve the Nash equilibrium, a stable strategy combination can be found in the conflicts and cooperation of multiple interests, so that each participant has no motivation to change the strategy individually under this strategy, thereby achieving a better overall operating state of the system and improving the scientific decision-making of virtual power plants in complex environments.

[0038] Collect the real-time status data of the virtual power plant, combine it with the equilibrium strategy set and risk gradient matrix of the virtual power plant to perform manifold optimization, and output the optimization instructions of the virtual power plant. The specific process is as follows: Collect real-time status data of the virtual power plant, including the voltage of the i-th node of the virtual power plant , virtual power plant line load rate ; Riemannian metric construction: ; In the formula, x is the strategy variable corresponding to the photovoltaic operator, y is the strategy variable corresponding to the energy storage system, and z is the strategy variable corresponding to the power grid. is the photovoltaic cost weight, is the energy storage life weight stored in the database, is the grid stability weight stored in the database, is the line load penalty term stored in the database, s is the arc length parameter on the Riemannian manifold; The geodesic equation is numerically solved using the fourth-order Runge-Kutta method: ; Where, is the risk gradient matrix, 、 、 is the tensor index, Corresponding power generation, Corresponding charge and discharge power, Corresponding to the grid interaction power, is the Christoffel symbol, t is the time index; Dynamic weight adjustment: ; Where, is the photovoltaic cost weight at time t, is the initial photovoltaic cost weight stored in the database, e is a natural constant, is the frequency deviation of the virtual power plant grid at time t; Optimization instructions for virtual power plants generate: ; Where, To optimize power generation, To optimize the charge and discharge power, To optimize grid interaction power.

[0039] Based on manifold learning theory, it can mine the inherent low-dimensional structure of real-time virtual power plant status data (such as node voltages and line load factors) within a high-dimensional space. Manifold learning effectively captures the complex nonlinear relationships between these data, making it more realistic than traditional Euclidean space analysis and providing a more accurate data foundation for subsequent optimization.

[0040] By constructing a Riemann metric, we consider multiple factors, including photovoltaic costs, energy storage lifespan, grid stability, and line load, and assign appropriate weights to each factor. This allows us to comprehensively weigh the impact of different factors on virtual power plant operations, avoid over-dominance of a single factor, and achieve more comprehensive and balanced scheduling optimization.

[0041] The PV cost weighting is dynamically adjusted based on grid frequency deviation, making weighting more flexible. Under different operating conditions, the PV cost weighting can be adaptively adjusted based on frequency deviation, better adapting to changes in the virtual power plant's operating state and optimizing resource allocation.

[0042] The fourth-order Runge-Kutta method is used to numerically solve the geodesic equation. This method is an effective means of solving ordinary differential equations. It can find the optimal trajectory of the system state variables in manifold space, providing a feasible numerical calculation method for optimizing virtual power plant scheduling strategies and making the optimization instructions practical.

[0043] Collect real-time mutation feature data of the virtual power plant. The specific process is as follows: Collect real-time mutation feature data of virtual power plants, including fault signs of virtual power plant equipment , Virtual Power Plant Wind Speed Sudden Change Signal , virtual power plant frequency secondary deviation ; Where n is the number of virtual power plant devices.

[0044] Based on the optimization instructions and real-time mutation feature data of the virtual power plant, quantum annealing fault tolerance is analyzed. The specific process is as follows: Construct the Hamiltonian H: ; Where, is the parameter of the interaction strength between device p and device q in the system, is the component of the Pauli matrix of device p in the B direction, is the component of the Pauli matrix of device q in the B direction, is the transverse field intensity at time t, is the component of the Pauli matrix of device p in the A direction, is the fault penalty coefficient stored in the database, is the fault flag of the pth device, p and q are the device numbers, and t is the time index; Tunneling probability calculate: ; Where, is the reduced Planck constant, is the transverse field intensity, is the energy difference; Obtain the tunneling probability-time constant mapping set stored in the database, and determine the matching time constant based on the tunneling probability .

[0045] And determine the dynamic adjustment rules and output the final execution signal of the virtual power plant. The specific process is as follows: Dynamic adjustment rules: Photovoltaic off-grid: ; Where, is the final charge and discharge power, To optimize the charge and discharge power, is the power variation stored in the database, e is a natural constant, and t is a time index; Frequency exceeds the limit: ; Where, is the power adjustment amount, is the scale factor stored in the database, is the integral coefficient stored in the database; Output the final execution signal of the virtual power plant: ; ; Where, To optimize power generation, is the final power generation; ; Where, To optimize grid interaction power, is the final grid interaction power.

[0046] By collecting real-time mutation feature data such as equipment fault signs, wind speed mutation signals, and frequency secondary deviations, it can timely perceive sudden changes in the operation of the virtual power plant, such as equipment failures and sudden changes in meteorological conditions, providing information support for rapid response and preventing small faults from turning into major accidents.

[0047] By constructing the Hamiltonian and calculating tunneling probabilities, and utilizing quantum annealing principles to analyze fault tolerance, this approach can explore the path of system state changes under complex mutations, identify effective strategies for addressing faults and other issues, and improve the stability and reliability of virtual power plants under abnormal operating conditions.

[0048] In the event of a PV system disconnection, the system dynamically adjusts the charge and discharge power based on optimized charge and discharge power, combined with power variation and time constant. This allows for reasonable adjustments to the energy storage charge and discharge power when the PV system is disconnected, maintaining the power balance of the virtual power plant, ensuring power supply reliability, and reducing impacts on the grid.

[0049] Based on the quadratic frequency deviation, the proportional-integral control rule is used to adjust power. This effectively addresses frequency overshoots, quickly adjusts grid interaction power, restores system frequency stability, and enhances the virtual power plant's ability to regulate grid frequency fluctuations.

[0050] After comprehensively considering various situations, clear execution signals such as the final power generation power, final charging and discharging power, and final grid interaction power are given to provide clear guidance for the actual operation of the virtual power plant, ensure the coordinated operation of various equipment, and realize the effective control and stable operation of the virtual power plant under complex working conditions.

[0051] Example 2: Figure 3 As shown, the artificial intelligence-based virtual power plant energy dispatching system is used to implement the method in Example 1, including: Standard LSTM operation module, used to collect multimodal features of virtual power plants and perform standard LSTM operations; The elasticity correction equation construction module is used to construct the elasticity correction equation, combine the correction coefficient of historical electricity price and load data, and output the virtual power plant power forecast value and elasticity correction amount; The equilibrium strategy set acquisition module is used to obtain the real-time characteristic data and physical constraints of the virtual power plant, and conduct a three-party dynamic game based on the virtual power plant power forecast value and elasticity correction value to obtain the equilibrium strategy set and risk gradient matrix of the virtual power plant; The optimization instruction analysis module is used to collect real-time status data of the virtual power plant, perform manifold optimization based on the equilibrium strategy set and risk gradient matrix of the virtual power plant, and output the optimization instructions of the virtual power plant; The final execution signal output module is used to collect the real-time mutation feature data of the virtual power plant, analyze the quantum annealing fault tolerance based on the optimization instructions and real-time mutation feature data of the virtual power plant, determine the dynamic adjustment rules, and output the final execution signal of the virtual power plant.

Claims

1. The energy dispatching method of virtual power plant based on artificial intelligence is characterized by: The following steps are involved: Collect multimodal features of the virtual power plant, perform standard LSTM operations, and determine the correction coefficient of historical electricity price and load data; Construct an elasticity correction equation, combine it with the correction coefficient of historical electricity price and load data, and output the virtual power plant power forecast value and elasticity correction amount; Acquire the real-time characteristic data and physical constraints of the virtual power plant, combine the virtual power plant power forecast value and elasticity correction value to conduct a three-party dynamic game, and obtain the equilibrium strategy set and risk gradient matrix of the virtual power plant; Collect real-time status data of the virtual power plant, perform manifold optimization based on the equilibrium strategy set and risk gradient matrix of the virtual power plant, and output optimization instructions for the virtual power plant; Collect the real-time mutation feature data of the virtual power plant, analyze the quantum annealing fault tolerance based on the optimization instructions and real-time mutation feature data of the virtual power plant, determine the dynamic adjustment rules, and output the final execution signal of the virtual power plant.

2. The method for energy scheduling of a virtual power plant based on artificial intelligence according to claim 1, characterized in that: The multimodal features of the virtual power plant are collected and standard LSTM operations are performed. The specific process is as follows: Get the hidden state of the LSTM at time t-1 stored in the database and the cell state at time t-1 ; Collect multimodal characteristics of virtual power plants, including photovoltaic power of virtual power plants , Virtual power plant fan speed , virtual power plant irradiance , virtual power plant wind turbine power ; Perform standard LSTM operations: Forget Gate: ; Where, is the forget gate output at time t, is the sigmoid activation function, is the forget gate weight matrix, is the input at the current time t, is the forget gate bias term; Input Gate: ; Where, is the input gate output at time t, is the input gate weight matrix, is the input gate bias term; Candidate cell states: ; Where, is the state output of the candidate cell at time t, is the candidate cell state weight matrix, is the candidate cell state bias; Cell status update: ; Where, is the cell state output at time t; Output gate: ; Where, is the output gate output at time t, is the output gate weight matrix, is the output gate bias term; Basic hidden state: ; Where, is the basic hidden state at time t; Physical gradient term injection: ； Where, is the physical gradient term, is the physical gradient term scaling factor stored in the database, Calibration factor for the IV curve of monocrystalline silicon modules stored in the database, is the fan characteristic curve fitting factor stored in the database; Final status update: ; ； Where, is the rate of change of the hidden layer state h over time, is the hidden state of LSTM at time t; Obtain the hidden state of LSTM stored in the database - the historical electricity price load data correction coefficient mapping set, and determine the matching historical electricity price load data correction coefficient based on the hidden state of LSTM at the current time t .

3. The method for energy scheduling of a virtual power plant based on artificial intelligence according to claim 2, characterized in that: The elasticity correction equation is constructed, combined with the correction coefficient of historical electricity price and load data, to output the virtual power plant power forecast value and elasticity correction amount. The specific process is as follows: Collect the power sequence of the virtual power plant in the past 24 hours and real-time electricity prices for virtual power plants ; LSTM basic prediction: ; Where, The power value of the virtual power plant at time t obtained by LSTM-based prediction; Calculation of elastic correction term: ; Where, is the elastic correction, is the demand price elasticity coefficient stored in the database, is the rate of change of the real-time electricity price of the virtual power plant over time; Final prediction output: ; Where, is the predicted power value of the virtual power plant at time t.

4. The method for energy scheduling of a virtual power plant based on artificial intelligence according to claim 1, characterized in that: The specific process of obtaining the real-time characteristic data and physical constraints of the virtual power plant is as follows: Obtain real-time characteristic data of the virtual power plant, including the electricity price of the i-th node of the virtual power plant , Virtual Power Plant Energy Storage SOC Status , virtual power plant grid frequency deviation , i is the virtual power plant node number; Capture the physical constraints of the virtual power plant, including: Obtain the virtual power plant irradiance-photovoltaic maximum output mapping set stored in the database, and determine the matching photovoltaic maximum output based on the current virtual power plant irradiance ; Virtual power plant wind turbine ramp rate constraints ; in, is the wind turbine power of the virtual power plant, is the rated power of the virtual power plant wind turbine, is the rate of change of wind turbine power in the virtual power plant over time; The physical constraints of the virtual power plant are used as constraints to obtain the equilibrium strategy set of the virtual power plant; The electricity price and maximum photovoltaic output of each node of the virtual power plant are stored as specified tags, and the specified tag-photovoltaic operation and maintenance cost mapping set stored in the database is obtained. Based on the current specified tag, the matching photovoltaic operation and maintenance cost is determined. .

5. The method for energy scheduling of a virtual power plant based on artificial intelligence according to claim 4 is characterized in that: The three-party dynamic game is conducted by combining the virtual power plant power forecast value and the elasticity correction value to obtain the equilibrium strategy set and risk gradient matrix of the virtual power plant. The specific process is as follows: Three-party dynamic game modeling: The participants are: PV operators, energy storage systems, and power grids; PV operator's corresponding strategy variable x: adjust the power generation ; Energy storage system corresponding strategy variable y: control charging and discharging power ; Grid corresponding strategy variable z: adjust the grid interaction power ; Strategy Space: ； Where, is the predicted power value of the virtual power plant at time t; ； Where, is the maximum discharge power stored in the database, The maximum charging power stored in the database; ； Where, is the minimum grid interaction power stored in the database, is the maximum grid interaction power stored in the database; Potential energy function : ； Where, is the potential energy function corresponding to the three strategy variables x, y, and z, is the photovoltaic operation and maintenance cost coefficient stored in the database, is the energy storage aging penalty coefficient stored in the database, is the first frequency sensitivity coefficient stored in the database, is the second frequency sensitivity coefficient stored in the database, and e is a natural constant; The ultimate goal is to minimize the total potential energy; Solve the Nash equilibrium and iteratively update the strategy until convergence: ； In the formula, k is the number of iterations, is the value of variable x after the k+1th iteration, is the value of variable y after the k+1th iteration, is the value of variable z after the k+1th iteration, is the value of variable x at the kth iteration, is the value of variable y at the kth iteration, is the value of variable z at the kth iteration, is the step length; After convergence, output the equilibrium strategy set of the virtual power plant ; is the target power generation, is the target charge and discharge power, is the target grid interaction power; Risk Gradient Matrix for: 。 6. The method for energy scheduling of a virtual power plant based on artificial intelligence according to claim 1, characterized in that: The real-time status data of the virtual power plant is collected, and the equilibrium strategy set and risk gradient matrix of the virtual power plant are combined to perform manifold optimization, and the optimization instructions of the virtual power plant are output. The specific process is as follows: Collect real-time status data of the virtual power plant, including the voltage of the i-th node of the virtual power plant , virtual power plant line load rate ; Riemannian metric construction: ； In the formula, x is the strategy variable corresponding to the photovoltaic operator, y is the strategy variable corresponding to the energy storage system, and z is the strategy variable corresponding to the power grid. is the photovoltaic cost weight, is the energy storage life weight stored in the database, is the grid stability weight stored in the database, is the line load penalty term stored in the database, s is the arc length parameter on the Riemannian manifold; The geodesic equation is numerically solved using the fourth-order Runge-Kutta method: ； Where, is the risk gradient matrix, 、 、 is the tensor index, Corresponding power generation, Corresponding charge and discharge power, Corresponding to the grid interaction power, is the Christoffel symbol, t is the time index; Dynamic weight adjustment: ; Where, is the photovoltaic cost weight at time t, is the initial photovoltaic cost weight stored in the database, e is a natural constant, is the frequency deviation of the virtual power plant grid at time t; Optimization instructions for virtual power plants generate: ; Where, To optimize power generation, To optimize the charge and discharge power, To optimize grid interaction power.

7. The method for energy dispatching of a virtual power plant based on artificial intelligence according to claim 1, characterized in that: The specific process of collecting real-time mutation characteristic data of the virtual power plant is as follows: Collect real-time mutation feature data of virtual power plants, including fault signs of virtual power plant equipment , Virtual Power Plant Wind Speed Sudden Change Signal , virtual power plant frequency secondary deviation ; Where n is the number of virtual power plant devices.

8. The method for energy dispatching of a virtual power plant based on artificial intelligence according to claim 7, characterized in that: The optimization instructions and real-time mutation feature data based on the virtual power plant are used to analyze quantum annealing fault tolerance. The specific process is as follows: Construct the Hamiltonian H: ； Where, is the parameter of the interaction strength between device p and device q in the system, is the component of the Pauli matrix of device p in the B direction, is the component of the Pauli matrix of device q in the B direction, is the transverse field intensity at time t, is the component of the Pauli matrix of device p in the A direction, is the fault penalty coefficient stored in the database, is the fault flag of the pth device, p and q are the device numbers, and t is the time index; Tunneling probability calculate: ； Where, is the reduced Planck constant, is the transverse field intensity, is the energy difference; Obtain the tunneling probability-time constant mapping set stored in the database, and determine the matching time constant based on the tunneling probability .

9. The method for energy scheduling of a virtual power plant based on artificial intelligence according to claim 8, characterized in that: The dynamic adjustment rules are determined and the final execution signal of the virtual power plant is output. The specific process is as follows: Dynamic adjustment rules: Photovoltaic off-grid: ; Where, is the final charge and discharge power, To optimize the charge and discharge power, is the power variation stored in the database, e is a natural constant, and t is a time index; Frequency exceeds the limit: ; Where, is the power adjustment amount, is the scale factor stored in the database, is the integral coefficient stored in the database; Output the final execution signal of the virtual power plant: ； ； Where, To optimize power generation, is the final power generation; ； Where, To optimize grid interaction power, is the final grid interaction power.

10. An artificial intelligence-based virtual power plant energy dispatching system, used to implement the method according to any one of claims 1 to 9, characterized in that: include: Standard LSTM operation module, used to collect multimodal features of virtual power plants and perform standard LSTM operations; The elasticity correction equation construction module is used to construct the elasticity correction equation, combine the correction coefficient of historical electricity price and load data, and output the virtual power plant power forecast value and elasticity correction amount; The equilibrium strategy set acquisition module is used to obtain the real-time characteristic data and physical constraints of the virtual power plant, and conduct a three-party dynamic game based on the virtual power plant power forecast value and elasticity correction value to obtain the equilibrium strategy set and risk gradient matrix of the virtual power plant; The optimization instruction analysis module is used to collect real-time status data of the virtual power plant, perform manifold optimization based on the equilibrium strategy set and risk gradient matrix of the virtual power plant, and output the optimization instructions of the virtual power plant; The final execution signal output module is used to collect the real-time mutation feature data of the virtual power plant, analyze the quantum annealing fault tolerance based on the optimization instructions and real-time mutation feature data of the virtual power plant, determine the dynamic adjustment rules, and output the final execution signal of the virtual power plant.

Images