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

Abstract

The present application belongs to the technical field of energy scheduling, and specifically discloses a virtual power plant energy scheduling method and system based on artificial intelligence, which comprises the following steps: collecting the multi-modal characteristics of the virtual power plant, constructing an elastic correction equation, correcting the coefficient in combination with historical electricity price load data, outputting the virtual power plant power prediction value and the elastic correction amount, obtaining the real-time characteristic data and physical constraints of the virtual power plant, conducting three-party dynamic game in combination with the virtual power plant power prediction value and the elastic correction amount, collecting the real-time state data of the virtual power plant, outputting the optimization instruction of the virtual power plant, based on the optimization instruction and the real-time mutation characteristic data of the virtual power plant, analyzing quantum annealing fault tolerance, determining the dynamic adjustment rule, and outputting the final execution signal of the virtual power plant. Through the comprehensive integration of the multi-modal characteristics and real-time data of the virtual power plant, the present application realizes high-precision power prediction, intelligent resource scheduling and dynamic strategy adjustment, and significantly improves the accuracy, flexibility and stability of energy management.

Description

Technical Field

[0001] This invention belongs to the field of energy dispatching technology, specifically relating to a virtual power plant energy dispatching method and system based on artificial intelligence. Background Technology

[0002] Renewable energy generation is characterized by intermittency and volatility. For example, wind power is affected by wind speed, and photovoltaic power is affected by the duration and intensity of sunlight. Large-scale integration poses challenges to the stable operation of the power system, requiring more intelligent dispatching methods to balance supply and demand. Virtual power plants can aggregate various energy resources such as distributed generation, energy storage, and controllable loads to achieve coordinated development of generation, grid, load, and storage.

[0003] However, existing energy management systems mostly adopt localized and decentralized management strategies, lacking cross-regional and cross-industry coordination and integration capabilities. This leads to problems such as energy supply and demand imbalances, low transmission efficiency, and insufficient utilization of energy storage resources. With the large-scale integration of renewable energy sources, how to effectively manage the intermittent and distributed characteristics of these energy sources and ensure the stability and security of energy supply has become an unresolved issue.

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

[0005] To address the shortcomings of existing technologies, this invention provides an artificial intelligence-based virtual power plant energy dispatching method and system. By integrating advanced artificial intelligence technologies such as Long Short-Term Memory (LSTM) networks, elastic correction mechanisms, three-party dynamic game theory, manifold optimization, and quantum annealing fault tolerance, it comprehensively integrates the multimodal characteristics and real-time data of the virtual power plant, achieving high-precision power prediction, intelligent resource dispatching, and dynamic strategy adjustment. This significantly improves the accuracy, flexibility, and stability of energy management and optimizes energy utilization efficiency.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] The energy dispatching method for virtual power plants based on artificial intelligence includes the following steps:

[0008] Collect multimodal characteristics of virtual power plants, perform standard LSTM calculations, and determine the correction coefficients for historical electricity price load data;

[0009] Construct an elastic correction equation, combine it with historical electricity price load data correction coefficients, and output the virtual power plant power prediction value and elastic correction amount;

[0010] Real-time characteristic data and physical constraints of the virtual power plant are obtained, and a three-party dynamic game is conducted by combining the power prediction value and elastic correction amount of the virtual power plant to obtain the equilibrium strategy set and risk gradient matrix of the virtual power plant.

[0011] Collect real-time status data of the virtual power plant, and perform manifold optimization by combining the virtual power plant's equilibrium strategy set and risk gradient matrix, and output the optimization instructions of the virtual power plant.

[0012] Collect real-time mutation characteristic data of the virtual power plant, analyze quantum annealing fault tolerance based on the optimization instructions of the virtual power plant and the real-time mutation characteristic data, determine the dynamic adjustment rules, and output the final execution signal of the virtual power plant.

[0013] Preferably, the process of collecting multimodal features of the virtual power plant and performing standard LSTM operations is as follows:

[0014] Retrieve the hidden state of the LSTM at time t-1 stored in the database. and cell state at time t-1 ;

[0015] Collect multimodal features of the virtual power plant, including the photovoltaic power of the virtual power plant. Virtual power plant fan speed Virtual power plant irradiance Virtual power plant fan power ;

[0016] Perform standard LSTM operations:

[0017] Forgotten Gate: ;

[0018] In the formula, The output of the forget gate at time t. It is the sigmoid activation function. Here is the forget gate weight matrix. This is the input at time t. Forget gate bias term;

[0019] Input Gate: ;

[0020] In the formula, The input gate output is at time t. The input gate weight matrix, For input gate bias terms;

[0021] Candidate cell status: ;

[0022] In the formula, Output the state of the candidate cell at time t. This is the candidate cell state weight matrix. This refers to the candidate cell state bias term;

[0023] Cell status update: ;

[0024] In the formula, Output the cell state at time t;

[0025] Output gate: ;

[0026] In the formula, The output of the gate at time t is the output of the gate. This is the output gate weight matrix. This is the output gate bias term;

[0027] Basic hidden state: ;

[0028] In the formula, Let t be the basic hidden state;

[0029] Physical gradient term injection:

[0030] ;

[0031] In the formula, For the physical gradient term, The scaling factor for the physical gradient terms stored in the database. The calibration factor for the IV curves of monocrystalline silicon modules stored in the database. The fitting factor for the wind turbine characteristic curves stored in the database;

[0032] Final status update: ;

[0033] ;

[0034] In the formula, Let h be the rate of change of the hidden layer state over time. Let be the hidden state of the LSTM at time t;

[0035] Retrieve the LSTM hidden state-historical electricity price load data correction coefficient mapping set stored in the database. Based on the LSTM hidden state at time t, determine the matching historical electricity price load data correction coefficients. .

[0036] Preferably, the specific process of constructing the elastic correction equation, combining it with historical electricity price load data correction coefficients, and outputting the virtual power plant power prediction value and elastic correction amount is as follows:

[0037] Collect the power sequence of the virtual power plant over the past 24 hours and virtual power plant real-time electricity price ;

[0038] LSTM Basic Prediction: ;

[0039] In the formula, The power value of the virtual power plant at time t is obtained from the LSTM-based prediction.

[0040] Calculation of the elasticity correction term: ;

[0041] In the formula, This is an elastic correction amount. The price elasticity of demand coefficient stored in the database. The rate of change of the real-time electricity price of the virtual power plant over time;

[0042] Final prediction output: ;

[0043] In the formula, Let t be the predicted power output of the virtual power plant.

[0044] Preferably, the specific process for obtaining the real-time feature data and physical constraints of the virtual power plant is as follows:

[0045] 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 , where i is the virtual power plant node number;

[0046] Obtain the physical constraints of the virtual power plant, including:

[0047] Retrieve the virtual power plant irradiance-maximum photovoltaic output mapping set stored in the database, and determine the matching maximum photovoltaic output based on the current virtual power plant irradiance. ;

[0048] Virtual power plant wind turbine ramp rate constraints ;

[0049] in, For the virtual power plant's wind turbine power, The rated power of the fan in the virtual power plant. The rate of change of the power of the virtual power plant's wind turbines over time;

[0050] The physical constraints of the virtual power plant serve as constraints for obtaining the equilibrium strategy set of the virtual power plant.

[0051] The electricity price and maximum photovoltaic output of each node in the virtual power plant are stored as specified tags. The specified tag-PV operation and maintenance cost mapping set stored in the database is retrieved. Based on the current specified tag, the matching PV operation and maintenance cost is determined. .

[0052] Preferably, the process of combining the predicted power value of the virtual power plant with the elastic correction amount to conduct a three-party dynamic game to obtain the equilibrium strategy set and risk gradient matrix of the virtual power plant is as follows:

[0053] Three-party dynamic game modeling:

[0054] The participants are: photovoltaic operators, energy storage systems, and power grids;

[0055] The corresponding strategy variable x for photovoltaic operators: adjusting power generation. ;

[0056] The corresponding strategy variable y for the energy storage system is the control of charging and discharging power. ;

[0057] Grid-related strategy variable z: Adjusting grid interaction power ;

[0058] Strategy Space:

[0059] ;

[0060] In the formula, Let be the predicted power output of the virtual power plant at time t;

[0061] ;

[0062] In the formula, The maximum discharge power stored in the database. The maximum charging power stored in the database;

[0063] ;

[0064] In the formula, The minimum grid interaction power stored in the database. The maximum grid interaction power stored in the database;

[0065] Potential energy function :

[0066] ;

[0067] In the formula, Let x, y, and z be the potential energy functions corresponding to the three policy variables. The photovoltaic operation and maintenance cost coefficients are stored in the database. The energy storage aging penalty coefficient is stored in the database. The first sensitivity coefficient for frequency stored in the database. The second sensitivity coefficient for frequency is stored in the database, and e is the natural constant;

[0068] The ultimate goal is to minimize the total potential energy;

[0069] Perform Nash equilibrium solution and iteratively update the policy until convergence:

[0070] ;

[0071] In the formula, k is the iteration number. Let x be the value of the variable x after the (k+1)th iteration. Let y be the value of the variable y after the (k+1)th iteration. Let z be the value of the variable z after the (k+1)th iteration. Let x be the value of the variable x in the k-th iteration. Let y be the value of the variable y in the k-th iteration. Let z be the value of the variable z in the k-th iteration. Step size;

[0072] After convergence, output the equilibrium strategy set of the virtual power plant. ; For the target power generation, For the target charge / discharge power, For target grid interaction power;

[0073] Risk gradient matrix for:

[0074] .

[0075] Preferably, the process of collecting real-time status data of the virtual power plant, combining it with the virtual power plant's equilibrium strategy set and risk gradient matrix for manifold optimization, and outputting optimization instructions for the virtual power plant is as follows:

[0076] 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 ;

[0077] Riemannian metric construction:

[0078] ;

[0079] In the formula, x represents the strategy variable corresponding to the photovoltaic operator, y represents the strategy variable corresponding to the energy storage system, and z represents the strategy variable corresponding to the power grid. As a weighting factor for photovoltaic costs, The energy storage lifetime weights are stored in the database. The power grid stability weights are stored in the database. Here, s represents the line load penalty term stored in the database, and s is the arc length parameter on the Riemannian manifold.

[0080] The geodesic equations were solved numerically using the fourth-order Runge-Kutta method:

[0081] ;

[0082] In the formula, For the risk gradient matrix, , , For tensor indices, Corresponding power generation capacity Corresponding charging and discharging power, Corresponding grid interaction power, For Christofel notation, t is the time index;

[0083] Dynamic weight adjustment: ;

[0084] In the formula, Let t be the photovoltaic cost weight. The initial photovoltaic cost weights are stored in the database, and e is the natural constant. The virtual power plant grid frequency deviation at time t;

[0085] Optimization instructions for virtual power plants generate: ;

[0086] In the formula, To optimize power generation, To optimize charging and discharging power, To optimize power grid interaction.

[0087] Preferably, the specific process for collecting real-time mutation characteristic data of the virtual power plant is as follows:

[0088] Collect real-time abrupt change characteristic data of the virtual power plant, including equipment fault indicators. Virtual power plant wind speed change signal Virtual power plant frequency second deviation ;

[0089] Where n is the number of virtual power plant devices.

[0090] Preferably, the process of analyzing quantum annealing fault tolerance based on the optimization instructions and real-time mutation characteristic data of the virtual power plant is as follows:

[0091] Construct the Hamiltonian H:

[0092] ;

[0093] In the formula, Let be a parameter representing the interaction strength between device p and device q in the system. Let p be the component of the device's Pauli matrix in the B direction. Let q be the component of the device's Pauli matrix in the B direction. Let be the transverse field intensity at time t. Let p be the component of the device's Pauli matrix in the A direction. The fault penalty coefficient is stored in the database. Let q be the fault flag for the p-th device, where p and q are device numbers and t is the time index;

[0094] Tunneling probability calculate:

[0095] ;

[0096] In the formula, To reduce Planck's constant, For transverse field strength, Energy difference;

[0097] Retrieve the tunneling probability-time constant mapping set stored in the database, and determine the matching time constant based on the tunneling probability. .

[0098] Preferably, the process of determining the dynamic adjustment rules and outputting the final execution signal of the virtual power plant is as follows:

[0099] Dynamic adjustment rules:

[0100] Photovoltaic grid disconnection: ;

[0101] In the formula, For the final charge and discharge power, To optimize charging and discharging power, The power change is stored in the database, where e is the natural constant and t is the time index;

[0102] Frequency exceeding the limit: ;

[0103] In the formula, For power adjustment amount, The proportional coefficients stored in the database. These are the integral coefficients stored in the database;

[0104] Output the final execution signal of the virtual power plant:

[0105] ;

[0106] ;

[0107] In the formula, To optimize power generation, This refers to the final power generation capacity.

[0108] ;

[0109] In the formula, To optimize grid interconnection power, This represents the final power exchange between the power grid and the grid.

[0110] An AI-based virtual power plant energy dispatching system, used to implement the aforementioned AI-based virtual power plant energy dispatching method, includes:

[0111] The standard LSTM computing module is used to collect multimodal features of a virtual power plant and perform standard LSTM operations.

[0112] The elastic correction equation construction module is used to construct elastic correction equations, combine historical electricity price load data with correction coefficients, and output virtual power plant power prediction values ​​and elastic correction amounts.

[0113] The equilibrium strategy set acquisition module is used to acquire real-time feature data and physical constraints of the virtual power plant, and to conduct a three-party dynamic game by combining the power prediction value and elastic correction amount of the virtual power plant to obtain the equilibrium strategy set and risk gradient matrix of the virtual power plant.

[0114] The optimization instruction analysis module is used to collect real-time status data of the virtual power plant, and perform manifold optimization by combining the virtual power plant's equilibrium strategy set and risk gradient matrix, and output the optimization instructions of the virtual power plant.

[0115] The final execution signal output module is used to collect real-time mutation characteristic data of the virtual power plant, analyze quantum annealing fault tolerance based on the optimization instructions of the virtual power plant and the real-time mutation characteristic data, determine the dynamic adjustment rules, and output the final execution signal of the virtual power plant.

[0116] The present invention has the following beneficial effects:

[0117] This invention 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 dynamic characteristics of distributed energy generation and load changes in a virtual power plant over time. For example, it captures fluctuations in photovoltaic output due to sunlight intensity and time, and the periodicity of user electricity consumption habits. Combining multimodal features allows for more comprehensive and accurate power prediction, providing a reliable basis for subsequent scheduling. An elastic correction equation is constructed to output the predicted power value and the elastic correction amount. Considering the uncertainties of new energy generation and load, the elastic correction can dynamically adjust the predicted value based on real-time data and model analysis, reducing prediction errors and making the scheduling plan more closely match actual operating conditions.

[0118] This invention combines multi-faceted information to conduct a three-way dynamic game. The virtual power plant involves multiple stakeholders, including the generation side, the grid side, and the user side, enabling the rational allocation of resources among these stakeholders and improving overall operational efficiency. Simultaneously, the risk gradient matrix helps assess and manage scheduling risks. Manifold optimization is performed based on real-time state data and the previously obtained strategy set. Manifold learning can uncover the inherent low-dimensional structure of the data, finding a better scheduling trajectory in the virtual power plant's state space, making optimization instructions more precise, improving energy utilization efficiency, and reducing unnecessary energy losses.

[0119] This invention analyzes quantum annealing fault tolerance and determines dynamic adjustment rules by combining optimized instructions and real-time mutation characteristic data. When faced with sudden changes in renewable energy generation due to extreme weather or unexpected failures, it can quickly respond and adjust scheduling strategies. Quantum annealing fault tolerance analysis ensures stable system operation under complex conditions, while dynamic adjustment rules provide flexibility and adaptability to ensure reliable power supply from the virtual power plant. Attached Figure Description

[0120] Figure 1 This is a schematic flowchart of the method of the present invention;

[0121] Figure 2 This is a detailed flowchart of the method of the present invention;

[0122] Figure 3 This is a schematic diagram of the module connections of the system of the present invention. Detailed Implementation

[0123] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0124] In this embodiment, the formulas can be dimensionless as needed during calculation to simplify the calculation.

[0125] Example 1: As Figure 1 , Figure 2As shown, the energy dispatching method for virtual power plants based on artificial intelligence includes the following steps:

[0126] Collect multimodal characteristics of virtual power plants, perform standard LSTM calculations, and determine the correction coefficients for historical electricity price load data;

[0127] Construct an elastic correction equation, combine it with historical electricity price load data correction coefficients, and output the virtual power plant power prediction value and elastic correction amount;

[0128] Real-time characteristic data and physical constraints of the virtual power plant are obtained, and a three-party dynamic game is conducted by combining the power prediction value and elastic correction amount of the virtual power plant to obtain the equilibrium strategy set and risk gradient matrix of the virtual power plant.

[0129] Collect real-time status data of the virtual power plant, and perform manifold optimization by combining the virtual power plant's equilibrium strategy set and risk gradient matrix, and output the optimization instructions of the virtual power plant.

[0130] Collect real-time mutation characteristic data of the virtual power plant, analyze quantum annealing fault tolerance based on the optimization instructions of the virtual power plant and the real-time mutation characteristic data, determine the dynamic adjustment rules, and output the final execution signal of the virtual power plant.

[0131] The process involves collecting multimodal features from a virtual power plant and performing standard LSTM operations.

[0132] Retrieve the hidden state of the LSTM at time t-1 stored in the database. and cell state at time t-1 ;

[0133] Collect multimodal features of the virtual power plant, including the photovoltaic power of the virtual power plant. Virtual power plant fan speed Virtual power plant irradiance Virtual power plant fan power ;

[0134] Perform standard LSTM operations:

[0135] Forgotten Gate: ;

[0136] In the formula, The output of the forget gate at time t. It is the sigmoid activation function. Here is the forget gate weight matrix. This is the input at time t. Forget gate bias term;

[0137] Input Gate: ;

[0138] In the formula, The input gate output is at time t. The input gate weight matrix, For input gate bias terms;

[0139] Candidate cell status: ;

[0140] In the formula, Output the state of the candidate cell at time t. This is the candidate cell state weight matrix. This refers to the candidate cell state bias term;

[0141] Cell status update: ;

[0142] In the formula, Output the cell state at time t;

[0143] Output gate: ;

[0144] In the formula, The output of the gate at time t is the output of the gate. This is the output gate weight matrix. This is the output gate bias term;

[0145] Basic hidden state: ;

[0146] In the formula, Let t be the basic hidden state;

[0147] Physical gradient term injection:

[0148] ;

[0149] In the formula, For the physical gradient term, The scaling factor for the physical gradient terms stored in the database. The calibration factor for the IV curves of monocrystalline silicon modules stored in the database. The fitting factor for the wind turbine characteristic curves stored in the database;

[0150] Final status update: ;

[0151] ;

[0152] In the formula, Let h be the rate of change of the hidden layer state over time. Let be the hidden state of the LSTM at time t;

[0153] Retrieve the LSTM hidden state-historical electricity price load data correction coefficient mapping set stored in the database. Based on the LSTM hidden state at time t, determine the matching historical electricity price load data correction coefficients. .

[0154] LSTM possesses memory capabilities and, through structures such as forget gates, input gates, and output gates, can effectively process time-series data. The forget gate determines which information from the previous cell state is discarded, the input gate controls the addition of current input information, and the output gate determines the output content. In virtual power plants, it can effectively capture the dynamic changes in energy data over time, such as the fluctuations in photovoltaic power and wind turbine power. Compared to traditional neural networks, it is better at handling long-term and short-term dependencies, improving prediction accuracy.

[0155] The system collects multimodal features from a virtual power plant, including photovoltaic power, wind turbine speed, irradiance, and wind turbine power. LSTM can comprehensively process these different types of interrelated features to fully reflect the operating status of the virtual power plant and uncover potential relationships between data, such as analyzing the mapping relationship between irradiance and photovoltaic power, thereby more accurately predicting power generation.

[0156] By injecting physical gradient terms and combining them with the physical characteristics related to photovoltaic power and wind turbine power (determined by the calibration factor of the IV curve of monocrystalline silicon modules and the fitting factor of wind turbine characteristic curves), the model not only learns from data but also incorporates physical knowledge and actual operating laws. This enhances the model's understanding and simulation capabilities of the complex physical processes of virtual power plants, reduces prediction bias, and improves model accuracy.

[0157] Constructing an elastic correction equation, combining historical electricity price load data with correction coefficients, and outputting the virtual power plant power prediction value and elastic correction amount, the specific process is as follows:

[0158] Collect the power sequence of the virtual power plant over the past 24 hours and virtual power plant real-time electricity price ;

[0159] LSTM Basic Prediction: ;

[0160] In the formula, The power value of the virtual power plant at time t is obtained from the LSTM-based prediction.

[0161] Calculation of the elasticity correction term: ;

[0162] In the formula, This is an elastic correction amount. The price elasticity of demand coefficient stored in the database. The rate of change of the real-time electricity price of the virtual power plant over time;

[0163] Final prediction output: ;

[0164] In the formula, Let t be the predicted power output of the virtual power plant.

[0165] We collected the past 24-hour power series data from a virtual power plant and performed LSTM-based forecasting. LSTM excels at processing time series data, uncovering patterns, periodicities, and trends in power data over time. For example, it can identify patterns in daily load curves and fluctuations in renewable energy generation, providing a solid foundation for power forecasting. Compared to simpler methods, it can more accurately capture dynamic changes in power.

[0166] Real-time electricity prices and their rate of change over time are incorporated into the calculation of the elasticity correction term. Electricity prices reflect the supply and demand relationship of electricity, and changes in real-time electricity prices can guide users to adjust their electricity consumption behavior, thereby affecting the power output of virtual power plants. By incorporating dynamic changes in electricity prices into the forecast, the model can capture the response relationship between electricity prices and power, making the forecast more closely reflect the actual market situation.

[0167] The LSTM base forecast results are adjusted using an elastic correction term. Considering that the operation of a virtual power plant is affected by various uncertainties, the elastic correction can adjust the base forecast in real time according to the dynamic changes in electricity prices, making up for the limitations of simply relying on historical power forecasts, reducing forecast errors, making the final forecast output closer to the actual power value, and improving forecast accuracy.

[0168] The specific process for obtaining real-time characteristic data and physical constraints of the virtual power plant is as follows:

[0169] 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 , where i is the virtual power plant node number;

[0170] Obtain the physical constraints of the virtual power plant, including:

[0171] Retrieve the virtual power plant irradiance-maximum photovoltaic output mapping set stored in the database, and determine the matching maximum photovoltaic output based on the current virtual power plant irradiance. ;

[0172] Virtual power plant wind turbine ramp rate constraints ;

[0173] in, For the virtual power plant's wind turbine power, The rated power of the fan in the virtual power plant. The rate of change of the power of the virtual power plant's wind turbines over time;

[0174] The physical constraints of the virtual power plant serve as constraints for obtaining the equilibrium strategy set of the virtual power plant.

[0175] The electricity price and maximum photovoltaic output of each node in the virtual power plant are stored as specified tags. The specified tag-PV operation and maintenance cost mapping set stored in the database is retrieved. Based on the current specified tag, the matching PV operation and maintenance cost is determined. .

[0176] By combining the predicted power output of the virtual power plant with the elastic correction, a three-way dynamic game is conducted to obtain the equilibrium strategy set and risk gradient matrix of the virtual power plant. The specific process is as follows:

[0177] Three-party dynamic game modeling:

[0178] The participants are: photovoltaic operators, energy storage systems, and power grids;

[0179] The corresponding strategy variable x for photovoltaic operators: adjusting power generation. ;

[0180] The corresponding strategy variable y for the energy storage system is the control of charging and discharging power. ;

[0181] Grid-related strategy variable z: Adjusting grid interaction power ;

[0182] Strategy Space:

[0183] ;

[0184] In the formula, Let be the predicted power output of the virtual power plant at time t;

[0185] ;

[0186] In the formula, The maximum discharge power stored in the database. The maximum charging power stored in the database;

[0187] ;

[0188] In the formula, The minimum grid interaction power stored in the database. The maximum grid interaction power stored in the database;

[0189] Potential energy function :

[0190] ;

[0191] In the formula, Let x, y, and z be the potential energy functions corresponding to the three policy variables. The photovoltaic operation and maintenance cost coefficients are stored in the database. The energy storage aging penalty coefficient is stored in the database. The first sensitivity coefficient for frequency stored in the database. The second sensitivity coefficient for frequency is stored in the database, and e is the natural constant;

[0192] The ultimate goal is to minimize the total potential energy;

[0193] Perform Nash equilibrium solution and iteratively update the policy until convergence:

[0194] ;

[0195] In the formula, k is the iteration number. Let x be the value of the variable x after the (k+1)th iteration. Let y be the value of the variable y after the (k+1)th iteration. Let z be the value of the variable z after the (k+1)th iteration. Let x be the value of the variable x in the k-th iteration. Let y be the value of the variable y in the k-th iteration. Let z be the value of the variable z in the k-th iteration. Step size;

[0196] After convergence, output the equilibrium strategy set of the virtual power plant. ; For the target power generation, For the target charge / discharge power, For target grid interaction power;

[0197] Risk gradient matrix for:

[0198] .

[0199] It acquires real-time characteristic data such as virtual power plant node electricity price, energy storage SOC status, and grid frequency deviation, covering electricity market price signals, energy storage device status, and key grid operation indicators. It can comprehensively reflect the operating environment and status of virtual power plants and provide rich information for formulating reasonable strategies.

[0200] Considering physical constraints such as the maximum output of photovoltaic power and the ramp-up rate of wind turbines, ensure that the dispatching strategy is formulated within the actual operating capacity of the virtual power plant equipment, avoid equipment damage or inability to execute due to over-dispatch, and ensure the safe and stable operation of the system.

[0201] The system introduces a dynamic game involving photovoltaic operators, energy storage systems, and the power grid. Different stakeholders have different interests and regulatory approaches; through this game, the actions of all parties can be coordinated, leading to a more rational allocation of resources such as photovoltaic power generation, energy storage charging and discharging, and grid interaction power, thereby improving the overall operational efficiency and economic benefits of the virtual power plant.

[0202] By comprehensively reflecting factors such as photovoltaic operation and maintenance costs, energy storage aging penalties, and frequency deviation impacts through the potential energy function, costs and risks of different natures are quantified into a unified indicator, which facilitates the evaluation and comparison of the overall cost of the system under different strategies and provides a clear target for strategy optimization.

[0203] By employing an iterative update strategy until convergence to solve for the Nash equilibrium, a stable strategy combination can be found amidst conflicts and cooperation among multiple stakeholders. This ensures that none of the participants have the incentive to change their strategies individually, thereby achieving a better overall system operation and improving the scientific nature of virtual power plant decision-making in complex environments.

[0204] The system collects real-time status data from the virtual power plant and performs manifold optimization by combining the virtual power plant's equilibrium strategy set and risk gradient matrix, outputting optimization instructions for the virtual power plant. The specific process is as follows:

[0205] 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 ;

[0206] Riemannian metric construction:

[0207] ;

[0208] In the formula, x represents the strategy variable corresponding to the photovoltaic operator, y represents the strategy variable corresponding to the energy storage system, and z represents the strategy variable corresponding to the power grid. As a weighting factor for photovoltaic costs, The energy storage lifetime weights are stored in the database. The power grid stability weights are stored in the database. Here, s represents the line load penalty term stored in the database, and s is the arc length parameter on the Riemannian manifold.

[0209] The geodesic equations were solved numerically using the fourth-order Runge-Kutta method:

[0210] ;

[0211] In the formula, For the risk gradient matrix, , , For tensor indices, Corresponding power generation capacity Corresponding charging and discharging power, Corresponding grid interaction power, For Christofel notation, t is the time index;

[0212] Dynamic weight adjustment: ;

[0213] In the formula, Let t be the photovoltaic cost weight. The initial photovoltaic cost weights are stored in the database, and e is the natural constant. The virtual power plant grid frequency deviation at time t;

[0214] Optimization instructions for virtual power plants generate: ;

[0215] In the formula, To optimize power generation, To optimize charging and discharging power, To optimize power grid interaction.

[0216] Based on manifold learning theory, it is possible to uncover the inherent low-dimensional structure of real-time state data (such as node voltage and line load rate) of virtual power plants in a high-dimensional space. These data exhibit complex nonlinear relationships, which manifold learning can effectively capture, providing a more realistic picture than traditional Euclidean space analysis and offering a more accurate data foundation for subsequent optimization.

[0217] By constructing a Riemann metric, considering factors such as photovoltaic cost, energy storage lifespan, grid stability, and line load, and assigning appropriate weights to each factor, the system can comprehensively weigh the impact of different factors on the operation of the virtual power plant, avoiding the excessive dominance of a single factor and achieving more comprehensive and balanced dispatch optimization.

[0218] The photovoltaic cost weight is dynamically adjusted according to the grid frequency deviation, making the weight setting more flexible. Under different operating conditions, the photovoltaic cost weight can be adaptively adjusted according to the frequency deviation, better adapting to changes in the virtual power plant's operating status and optimizing resource allocation.

[0219] The fourth-order Runge-Kutta method is used to numerically solve the geodesic equations. This method is an effective means of solving ordinary differential equations, and can find the optimal trajectory of the system state variables in the manifold space. It provides a feasible numerical calculation approach for optimizing virtual power plant dispatch strategies, making the optimization instructions practically operable.

[0220] The process of collecting real-time mutation characteristic data of the virtual power plant is as follows:

[0221] Collect real-time abrupt change characteristic data of the virtual power plant, including equipment fault indicators. Virtual power plant wind speed change signal Virtual power plant frequency second deviation ;

[0222] Where n is the number of virtual power plant devices.

[0223] Based on the optimization instructions and real-time mutation characteristic data of the virtual power plant, the fault tolerance of quantum annealing is analyzed. The specific process is as follows:

[0224] Construct the Hamiltonian H:

[0225] ;

[0226] In the formula, Let be a parameter representing the interaction strength between device p and device q in the system. Let p be the component of the device's Pauli matrix in the B direction. Let q be the component of the device's Pauli matrix in the B direction. Let be the transverse field intensity at time t. Let p be the component of the device's Pauli matrix in the A direction. The fault penalty coefficient is stored in the database. Let q be the fault flag for the p-th device, where p and q are device numbers and t is the time index;

[0227] Tunneling probability calculate:

[0228] ;

[0229] In the formula, To reduce Planck's constant, For transverse field strength, Energy difference;

[0230] Retrieve the tunneling probability-time constant mapping set stored in the database, and determine the matching time constant based on the tunneling probability. .

[0231] The dynamic adjustment rules are determined, and the final execution signal of the virtual power plant is output. The specific process is as follows:

[0232] Dynamic adjustment rules:

[0233] Photovoltaic grid disconnection: ;

[0234] In the formula, For the final charge and discharge power, To optimize charging and discharging power, The power change is stored in the database, where e is the natural constant and t is the time index;

[0235] Frequency exceeding the limit: ;

[0236] In the formula, For power adjustment amount, The proportional coefficients stored in the database. These are the integral coefficients stored in the database;

[0237] Output the final execution signal of the virtual power plant:

[0238] ;

[0239] ;

[0240] In the formula, To optimize power generation, This refers to the final power generation capacity.

[0241] ;

[0242] In the formula, To optimize grid interconnection power, This represents the final power exchange between the power grid and the grid.

[0243] By collecting real-time data on sudden changes such as equipment fault signs, wind speed change signals, and frequency quadratic deviation, the system can promptly detect sudden changes in the operation of the virtual power plant, such as equipment failures and sudden changes in weather conditions, providing information support for rapid response and preventing minor faults from escalating into major accidents.

[0244] By constructing a Hamiltonian and calculating the tunneling probability, fault tolerance is analyzed using the principle of quantum annealing. This allows for the exploration of system state change paths under complex abrupt changes, identifying effective strategies to address faults and other problems, and improving the stability and reliability of virtual power plants under abnormal operating conditions.

[0245] In response to photovoltaic grid disconnection, dynamic adjustments are made based on optimized charging and discharging power, combined with power changes and time constants. This allows for reasonable adjustment of energy storage charging and discharging power when the photovoltaic system is disconnected from the grid, maintaining the power balance of the virtual power plant, ensuring power supply reliability, and reducing the impact on the power grid.

[0246] Based on the second-order frequency deviation, power is adjusted using proportional-integral control rules. This effectively addresses frequency over-limit issues, rapidly adjusts grid interaction power, restores system frequency stability, and enhances the virtual power plant's ability to regulate grid frequency fluctuations.

[0247] After comprehensively considering various situations, the system clearly provides execution signals such as the final power generation, final charging and discharging power, and final grid interaction power, providing clear guidance for the actual operation of the virtual power plant, ensuring the coordinated operation of all equipment, and achieving effective control and stable operation of the virtual power plant under complex operating conditions.

[0248] Example 2: Figure 3As shown, an artificial intelligence-based virtual power plant energy dispatching system, used to implement the method in Example 1, includes:

[0249] The standard LSTM computing module is used to collect multimodal features of a virtual power plant and perform standard LSTM operations.

[0250] The elastic correction equation construction module is used to construct elastic correction equations, combine historical electricity price load data with correction coefficients, and output virtual power plant power prediction values ​​and elastic correction amounts.

[0251] The equilibrium strategy set acquisition module is used to acquire real-time feature data and physical constraints of the virtual power plant, and to conduct a three-party dynamic game by combining the power prediction value and elastic correction amount of the virtual power plant to obtain the equilibrium strategy set and risk gradient matrix of the virtual power plant.

[0252] The optimization instruction analysis module is used to collect real-time status data of the virtual power plant, and perform manifold optimization by combining the virtual power plant's equilibrium strategy set and risk gradient matrix, and output the optimization instructions of the virtual power plant.

[0253] The final execution signal output module is used to collect real-time mutation characteristic data of the virtual power plant, analyze quantum annealing fault tolerance based on the optimization instructions of the virtual power plant and the real-time mutation characteristic data, determine the dynamic adjustment rules, and output the final execution signal of the virtual power plant.

Claims

1. An artificial intelligence-based virtual power plant energy dispatching method, characterized in that, Includes the following steps: Collect multimodal characteristics of virtual power plants, perform standard LSTM calculations, and determine the correction coefficients for historical electricity price load data; Construct an elastic correction equation, combine it with historical electricity price load data correction coefficients, and output the virtual power plant power prediction value and elastic correction amount; Real-time characteristic data and physical constraints of the virtual power plant are obtained, and a three-party dynamic game is conducted by combining the power prediction value and elastic correction amount of the virtual power plant to obtain the equilibrium strategy set and risk gradient matrix of the virtual power plant. The system collects real-time status data from the virtual power plant and performs manifold optimization by combining the virtual power plant's equilibrium strategy set and risk gradient matrix, outputting optimization instructions for 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 represents the strategy variable corresponding to the photovoltaic operator, y represents the strategy variable corresponding to the energy storage system, and z represents the strategy variable corresponding to the power grid. As a weighting factor for photovoltaic costs, The energy storage lifetime weights are stored in the database. The power grid stability weights are stored in the database. Here, s represents the line load penalty term stored in the database, and s is the arc length parameter on the Riemannian manifold. The geodesic equations were solved numerically using the fourth-order Runge-Kutta method: ； In the formula, For the risk gradient matrix, , , For tensor indices, Corresponding power generation capacity Corresponding charging and discharging power, Corresponding grid interaction power, For Christofel notation, t is the time index; Dynamic weight adjustment: ; In the formula, Let t be the photovoltaic cost weight. The initial photovoltaic cost weights are stored in the database, and e is the natural constant. The virtual power plant grid frequency deviation at time t; Optimization instructions for virtual power plants generate: ; In the formula, To optimize power generation, To optimize charging and discharging power, To optimize grid interconnection power; Collect real-time abrupt change characteristic data of the virtual power plant, including equipment fault indicators. Virtual power plant wind speed change signal Virtual power plant frequency second deviation ; Where n is the number of virtual power plant devices; Based on the optimization instructions and real-time mutation characteristic data of the virtual power plant, the fault tolerance of quantum annealing is analyzed. The specific process is as follows: Construct the Hamiltonian H: ； In the formula, Let be a parameter representing the interaction strength between device p and device q in the system. Let p be the component of the device's Pauli matrix in the B direction. Let q be the component of the device's Pauli matrix in the B direction. Let be the transverse field intensity at time t. Let p be the component of the device's Pauli matrix in the A direction. The fault penalty coefficient is stored in the database. Let q be the fault flag for the p-th device, where p and q are device numbers and t is the time index; Tunneling probability calculate: ； In the formula, To reduce Planck's constant, For transverse field strength, Energy difference; Retrieve the tunneling probability-time constant mapping set stored in the database, and determine the matching time constant based on the tunneling probability. ; It also determines the dynamic adjustment rules and outputs the final execution signal of the virtual power plant.

2. The energy dispatching method for virtual power plants based on artificial intelligence according to claim 1, characterized in that: The process of collecting multimodal features from a virtual power plant and performing standard LSTM operations is as follows: Retrieve the hidden state of the LSTM at time t-1 stored in the database. and cell state at time t-1 ; Collect multimodal features of the virtual power plant, including the photovoltaic power of the virtual power plant. Virtual power plant fan speed Virtual power plant irradiance Virtual power plant fan power ; Perform standard LSTM operations: Forgotten Gate: ; In the formula, The output of the forget gate at time t. It is the sigmoid activation function. Here is the forget gate weight matrix. This is the input at time t. Forget gate bias term; Input Gate: ; In the formula, The input gate output is at time t. The input gate weight matrix, For input gate bias terms; Candidate cell status: ; In the formula, Output the state of the candidate cell at time t. This is the candidate cell state weight matrix. This refers to the candidate cell state bias term; Cell status update: ; In the formula, Output the cell state at time t; Output gate: ; In the formula, The output of the gate at time t is the output of the gate. This is the output gate weight matrix. This is the output gate bias term; Basic hidden state: ; In the formula, Let t be the basic hidden state; Physical gradient term injection: ； In the formula, For the physical gradient term, The scaling factor for the physical gradient terms stored in the database. The calibration factor for the IV curves of monocrystalline silicon modules stored in the database. The fitting factor for the wind turbine characteristic curves stored in the database; Final status update: ; ； In the formula, Let h be the rate of change of the hidden layer state over time. Let be the hidden state of the LSTM at time t; Retrieve the LSTM hidden state-historical electricity price load data correction coefficient mapping set stored in the database. Based on the LSTM hidden state at time t, determine the matching historical electricity price load data correction coefficients. .

3. The energy dispatching method for virtual power plants based on artificial intelligence according to claim 2, characterized in that: The process of constructing the elastic correction equation, combining it with historical electricity price load data correction coefficients, and outputting the virtual power plant power prediction value and elastic correction amount is as follows: Collect the power sequence of the virtual power plant over the past 24 hours and virtual power plant real-time electricity price ; LSTM Basic Prediction: ; In the formula, The power value of the virtual power plant at time t is obtained from the LSTM-based prediction. Calculation of the elasticity correction term: ; In the formula, This is an elastic correction amount. The price elasticity of demand coefficient stored in the database. The rate of change of the real-time electricity price of the virtual power plant over time; Final prediction output: ; In the formula, Let t be the predicted power output of the virtual power plant.

4. The energy dispatching method for virtual power plants based on artificial intelligence according to claim 1, characterized in that: The specific process for obtaining the real-time feature 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 , where i is the virtual power plant node number; Obtain the physical constraints of the virtual power plant, including: Retrieve the virtual power plant irradiance-maximum photovoltaic output mapping set stored in the database, and determine the matching maximum photovoltaic output based on the current virtual power plant irradiance. ; Virtual power plant wind turbine ramp rate constraints ; in, For the virtual power plant's wind turbine power, The rated power of the fan in the virtual power plant. The rate of change of the power of the virtual power plant's wind turbines over time; The physical constraints of the virtual power plant serve as constraints for obtaining the equilibrium strategy set of the virtual power plant. The electricity price and maximum photovoltaic output of each node in the virtual power plant are stored as specified tags. The specified tag-PV operation and maintenance cost mapping set stored in the database is retrieved. Based on the current specified tag, the matching PV operation and maintenance cost is determined. .

5. The energy dispatching method for virtual power plants based on artificial intelligence according to claim 4, characterized in that: The process involves combining the predicted power output of the virtual power plant with the elastic correction value to conduct a three-way dynamic game, thereby obtaining the equilibrium strategy set and risk gradient matrix of the virtual power plant. The specific steps are as follows: Three-party dynamic game modeling: The participants are: photovoltaic operators, energy storage systems, and power grids; The corresponding strategy variable x for photovoltaic operators: adjusting power generation. ; The corresponding strategy variable y for the energy storage system is the control of charging and discharging power. ; Grid-related strategy variable z: Adjusting grid interaction power ; Strategy Space: ； In the formula, Let be the predicted power output of the virtual power plant at time t; ； In the formula, The maximum discharge power stored in the database. The maximum charging power stored in the database; ； In the formula, The minimum grid interaction power stored in the database. The maximum grid interaction power stored in the database; Potential energy function : ； In the formula, Let x, y, and z be the potential energy functions corresponding to the three policy variables. The photovoltaic operation and maintenance cost coefficients are stored in the database. The energy storage aging penalty coefficient is stored in the database. The first sensitivity coefficient for frequency stored in the database. The second sensitivity coefficient for frequency is stored in the database, and e is the natural constant; The ultimate goal is to minimize the total potential energy; Perform Nash equilibrium solution and iteratively update the policy until convergence: ； In the formula, k is the iteration number. Let x be the value of the variable x after the (k+1)th iteration. Let y be the value of the variable y after the (k+1)th iteration. Let z be the value of the variable z after the (k+1)th iteration. Let x be the value of the variable x in the k-th iteration. Let y be the value of the variable y in the k-th iteration. Let z be the value of the variable z in the k-th iteration. Step size; After convergence, output the equilibrium strategy set of the virtual power plant. ; For the target power generation, For the target charge / discharge power, For target grid interaction power; Risk gradient matrix for: 。 6. The energy dispatching method for virtual power plants based on artificial intelligence according to claim 1, characterized in that: The process involves determining the dynamic adjustment rules and outputting the final execution signal for the virtual power plant. Dynamic adjustment rules: Photovoltaic grid disconnection: ; In the formula, For the final charge and discharge power, To optimize charging and discharging power, The power change is stored in the database, where e is the natural constant and t is the time index; Frequency exceeding the limit: ; In the formula, For power adjustment amount, The proportional coefficients stored in the database. These are the integral coefficients stored in the database; Output the final execution signal of the virtual power plant: ； ； In the formula, To optimize power generation, This refers to the final power generation capacity. ； In the formula, To optimize grid interconnection power, This represents the final power exchange between the power grid and the grid.

7. An artificial intelligence-based virtual power plant energy dispatching system, used to implement the method described in any one of claims 1-6, characterized in that, include: The standard LSTM computing module is used to collect multimodal features of a virtual power plant and perform standard LSTM operations. The elastic correction equation construction module is used to construct elastic correction equations, combine historical electricity price load data with correction coefficients, and output virtual power plant power prediction values ​​and elastic correction amounts. The equilibrium strategy set acquisition module is used to acquire real-time feature data and physical constraints of the virtual power plant, and to conduct a three-party dynamic game by combining the power prediction value and elastic correction amount of the virtual power plant 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, and perform manifold optimization by combining the virtual power plant's equilibrium strategy set and risk gradient matrix, and output the optimization instructions of the virtual power plant. The final execution signal output module is used to collect real-time mutation characteristic data of the virtual power plant, analyze quantum annealing fault tolerance based on the optimization instructions of the virtual power plant and the real-time mutation characteristic data, determine the dynamic adjustment rules, and output the final execution signal of the virtual power plant.

Images