Virtual power plant optimization scheduling system and method based on block chain

Abstract

The invention relates to the technical field of power system dispatching, in particular to a virtual power plant optimal dispatching system and method based on a block chain, and aims to solve the problem that in the prior art, high-precision prediction of the generating capacity and load of a virtual power plant cannot be realized by using an LSTM (Long Short Term Memory) to capture a time sequence long-term dependency relationship. Input features and the number of LSTM units cannot be adjusted to adapt to different scenes, and virtual power plant operation and resource configuration are not facilitated; according to the invention, high-precision prediction of the generating capacity and the load of the virtual power plant is realized by using an LSTM algorithm through an energy load joint prediction module, the long-term dependency relationship of a time sequence is captured by using an LSTM to improve the prediction accuracy, input features and the number of LSTM units can be adjusted to adapt to different scenes, multi-strategy guarantee convergence is adopted for training to prevent overfitting, and the prediction accuracy is improved. And finally outputting a key prediction value, thereby facilitating virtual power plant operation and resource configuration.

Description

Technical Field

[0001] The present invention relates to the technical field of power system dispatching, and more specifically, to a blockchain-based virtual power plant optimization dispatching system and method. Background Art

[0002] As the world's attention to renewable energy and smart grids gradually deepens, virtual power plants, as systems that integrate distributed power sources, energy storage equipment and adjustable loads, are playing an increasingly important role in the power energy system. However, traditional virtual power plant dispatching systems have many problems, such as high dispatching and operating costs, poor economic efficiency, lack of transparency, loss of trust, risk of private data leakage, and limited types of energy integration. These problems have hindered the further development of virtual power plants.

[0003] The patent application with reference publication number CN117726127A discloses a blockchain-based virtual power plant intelligent analysis and scheduling system, including: a scenario set construction module for collecting historical weather data and real-time weather data, and creating a scenario set based on the real-time weather data and historical weather data; a power module for calculating the weather data in the scenario set through a similarity algorithm to obtain weather data after deduplication, and performing power calculation on the deduplicated weather data to obtain the maximum power and minimum power under each weather data; a cost module for calculating the total power generation cost set based on the maximum power and minimum power; a scheduling module for generating scheduling instructions based on the total power generation cost set; a blockchain module for storing data generated by the scenario set construction module, the power module, the cost module, and the scheduling module and transmitting the scheduling instructions. The present invention solves the problems of low energy scheduling accuracy and data security risks in the prior art;

[0004] However, the above-mentioned reference patent reduces the impact of volatility through scenario expansion and data simulation, and uses blockchain to ensure secure data transmission and real-time updates to improve scheduling accuracy, but cannot use LSTM to capture long-term dependencies in time series to achieve high-precision prediction of virtual power plant power generation and load, and cannot adjust input features and the number of LSTM units to adapt to different scenarios, which is not conducive to virtual power plant operations and resource allocation; at the same time, it cannot accurately predict and make intelligent decisions to optimize the scheduling plan of the virtual power plant, and cannot reduce operating costs and carbon emissions while ensuring system energy balance and equipment safety. It cannot improve the utilization efficiency of renewable energy, and reduces the reliability and stability of the power supply system.

[0005] To this end, we propose a blockchain-based virtual power plant optimization scheduling system and method to address the above problems. Summary of the Invention

[0006] The purpose of the present invention is to provide a virtual power plant optimization scheduling system and method based on blockchain, which solves the problems that the existing technology cannot use LSTM to capture the long-term dependencies of time series to achieve high-precision prediction of virtual power plant power generation and load, cannot adjust input features and the number of LSTM units to adapt to different scenarios, and is not conducive to virtual power plant operation and resource allocation; at the same time, it cannot accurately predict and make intelligent decisions to optimize the scheduling plan of the virtual power plant, cannot reduce operating costs and reduce carbon emissions on the basis of ensuring system energy balance and equipment safety, cannot improve the utilization efficiency of renewable energy, and reduces the reliability and stability of the power supply system.

[0007] The purpose of the present invention is achieved through the following technical solutions:

[0008] A blockchain-based virtual power plant optimization and dispatching system, applied to the virtual power plant management platform, includes:

[0009] The data acquisition module is used to collect real-time operating data of the virtual power plant, including photovoltaic device power generation, wind turbine power generation, energy storage device discharge power, photovoltaic panel irradiation intensity, wind turbine wind speed, energy storage device remaining power, load-side power consumption, ambient temperature, grid voltage, and grid frequency;

[0010] The energy load joint prediction module is used to build a joint prediction model based on the historical operation data of the virtual power plant using the LSTM algorithm. The joint prediction model can predict the power generation of each distributed energy source in the virtual power plant and the load demand of the entire virtual power plant in the future.

[0011] The scheduling optimization module is used to build an intelligent scheduling optimization model based on the prediction results of the joint prediction model by introducing reinforcement learning and multi-objective evolutionary algorithms. The model is used to optimize the distributed energy generation parameters and generate the optimal scheduling plan for the virtual power plant operation.

[0012] The scheduling execution module is used to control the corresponding execution agencies to execute control measures according to the generated operation scheduling plan.

[0013] As a preferred embodiment of the present invention, the process of the energy load joint forecasting module processing the historical operation data of the virtual power plant includes:

[0014] Obtain historical operating data of the virtual power plant and perform preprocessing operations on the historical operating data of the virtual power plant;

[0015] Convert the pre-processed virtual power plant operation data into a time series form, treat the data of each time step as a vector, set a time window, and treat the data within the window as an input sequence;

[0016] The time series data is divided into training set, validation set and test set in chronological order, with the allocation ratio of 70% training, 15% validation and 15% test.

[0017] As a preferred embodiment of the present invention, the process of constructing a joint prediction model by the energy load joint prediction module includes:

[0018] Construct a joint prediction model to predict the power generation of each distributed energy source in the virtual power plant and the load demand of the entire virtual power plant over a period of time in the future. The model structure includes the following parts:

[0019] Input layer: Determine the input layer dimension p to be 10, and construct the input vector x by combining the data of each time step, x = [x1, x2, ..., x t …, x m ], where x t ∈R p is the input vector at time t, R p Represents a p-dimensional real number space, taking x as input data;

[0020] LSTM layer: The LTSM layer contains m LTSM units. At each time step t, the input data x will be processed in parallel by m LSTM units. The output dimension of each LSTM unit is d.

[0021] Fully connected layer: receives the output from the LSTM layer Map it to a 4-dimensional output vector Calculated using the following formula:

[0022]

[0023] Where W fc is the weight matrix of the fully connected layer, b fc is the bias vector of the fully connected layer, z is the linear transformation result of the fully connected layer, and f(·) is the activation function;

[0024] Output layer: directly outputs the prediction vector The predicted values ​​of photovoltaic power generation, wind turbine power generation, energy storage device discharge, and total load are expressed as follows:

[0025]

[0026] As a preferred embodiment of the present invention, the process of processing input data by the LSTM layer in the joint prediction model includes:

[0027] For each time step t, the computation process of the m-th LSTM unit is as follows:

[0028] If m=1, the input is x i , otherwise the input is the output of the previous layer (m-1) Calculated by the following formula:

[0029]

[0030] Where σ represents the sigmoid activation function, ⊙ represents element-wise multiplication, are the weight matrices of the input gate, forget gate, cell state, and output gate, respectively. are the bias vectors of the input gate, forget gate, cell state, and output gate, respectively.

[0031] As a preferred embodiment of the present invention, the training steps for constructing the joint prediction model of the energy load joint prediction module are as follows:

[0032] T1: Randomly initialize all weight matrices and bias vectors;

[0033] T2: Input the training data into the joint prediction model to perform calculations on the input layer, LSTM layer, fully connected layer, and output layer to obtain the prediction vector

[0034] T3: Use mean square error as the loss function: Where N is the number of training samples, y t is a real-valued vector;

[0035] T4: Use the backpropagation algorithm to calculate the gradient of the loss function with respect to all model parameters;

[0036] T5: Use the optimizer to update the model parameters and adjust the weight matrix and bias vector according to the calculated gradient;

[0037] T6: Repeat steps T2-T5 until the model converges or the preset stopping condition is reached.

[0038] As a preferred embodiment of the present invention, the process of introducing reinforcement learning and multi-objective evolutionary algorithm into the scheduling optimization module to construct an intelligent scheduling optimization model includes:

[0039] Get the prediction result of the joint prediction model. The prediction result of the joint prediction model is the prediction vector

[0040] The distributed energy generation parameters are defined as vector r = [P gf , P fl , P cl ] T , where P gf is the power generated by the photovoltaic device, P flis the power generated by the wind turbine, P cl Discharge power for energy storage equipment;

[0041] The model is built through the following multi-objective function, with the goals of minimizing the total comprehensive operating cost, minimizing carbon emissions and maximizing the utilization of renewable energy;

[0042] The virtual power plant cost function is as follows:

[0043] Among them C zy is the total operating cost, C gm is the unit price of electricity purchased from the power grid, C cj is the energy storage charging price, P gm,t is the amount of electricity purchased from the grid, which represents the power purchased from the grid at time step t, is the energy storage charging power, which represents the power of the energy storage device charged from the grid at time step t;

[0044] The carbon emission function of a virtual power plant is as follows:

[0045] Among them E zy is the total carbon emissions, γ is the carbon emission coefficient;

[0046] The renewable energy utilization rate function of the virtual power plant is as follows:

[0047] Among them U nl is the utilization rate of renewable energy;

[0048] The energy balance constraint is as follows: P gf,t +P fl,t +P cl,t +P gm,t =D xz,t ;

[0049] The power generation constraints are as follows:

[0050] The operating constraints of energy storage equipment are as follows:

[0051]

[0052] SOC min ≤SOC t ≤SOC max ;

[0053] Among them, SOC t Represents the remaining power of the energy storage device at time step t, η c Represents the charging efficiency of energy storage equipment, η d Indicates the discharge efficiency of the energy storage device, Represents the discharge power of the energy storage device, Pcl,max Indicates the maximum output power of the energy storage device, SOC min Indicates the minimum power of the energy storage device, SOC max Indicates the maximum capacity of the energy storage device;

[0054] Designing a reinforcement learning framework:

[0055] State space S t for:

[0056] Action space A t For: A t =[P gf,t ,P fl,t ,P cl,t ];

[0057] Reward function R t For: R t =-(αC zy,t +βE zy,t -λU nl,t ), where α, β, and λ are weight coefficients.

[0058] As a preferred embodiment of the present invention, the process of the scheduling optimization module optimizing the distributed energy generation parameters and generating the optimal scheduling plan for the virtual power plant operation includes:

[0059] The multi-objective evolutionary algorithm solves the Pareto optimal scheduling strategy as follows: the action sequence generated by reinforcement learning is used as the initial population and input into the multi-objective evolutionary algorithm for further optimization. The specific steps are as follows:

[0060] R1: Encoding scheduling scheme is chromosome

[0061] R2: Construct the objective function vector: F(x) = [C zy (d),E zy (d),-U nl (d)];

[0062] R3: Generate new scheduling strategy combinations using selection, crossover, and mutation operations;

[0063] R4: Select the non-inferior frontier solution set by Pareto superiority sorting

[0064] According to the scheduling priority, the solution set is selected from the non-inferior frontier Select the optimal scheduling time series solution:

[0065] As a preferred embodiment of the present invention, the process of the scheduling execution module controlling the corresponding execution mechanism to execute the control measures according to the generated operation scheduling plan includes:

[0066] After receiving the operation scheduling plan, the scheduling execution module immediately sends a control instruction to the corresponding execution agency. The execution agency executes the corresponding control measures according to the control instruction. The specific contents of the control measures are as follows:

[0067] Adjust the output of the photovoltaic inverter to adjust the power generated by the photovoltaic device from the current value to the optimal value;

[0068] By adjusting the angle or speed of the wind turbine blades, the wind turbine generator power can be adjusted to the optimal value;

[0069] Control the charging and discharging process of the energy storage device so that the discharge power of the energy storage device reaches the optimal value while ensuring that the battery health status is not affected.

[0070] As a preferred embodiment of the present invention, a blockchain-based virtual power plant optimization scheduling method includes the following steps:

[0071] Step 1: Real-time collection of virtual power plant operating data, including photovoltaic device power generation, wind turbine power generation, energy storage device discharge power, photovoltaic panel irradiation intensity, wind turbine wind speed, energy storage device remaining power, load-side power consumption, ambient temperature, grid voltage, and grid frequency;

[0072] Step 2: Based on the historical operating data of the virtual power plant, a joint prediction model is constructed using the LSTM algorithm. This joint prediction model can predict the power generation of each distributed energy source in the virtual power plant and the load demand of the entire virtual power plant over a period of time in the future.

[0073] Step 3: Based on the prediction results of the joint prediction model, reinforcement learning and multi-objective evolutionary algorithms are introduced to build an intelligent scheduling optimization model. The model is used to optimize the distributed energy generation parameters and generate the optimal scheduling plan for the virtual power plant operation.

[0074] Step 4: Control the corresponding actuators to execute control measures according to the generated operation scheduling plan. Compared with the prior art, the advantages of the present invention are:

[0075] (1) In the present invention, the LSTM algorithm is used to achieve high-precision prediction of the power generation and load of the virtual power plant through the energy load joint prediction module. The LSTM is used to capture the long-term dependency of the time series to improve the prediction accuracy. The input features and the number of LSTM units can be adjusted to adapt to different scenarios. The training adopts multiple strategies to ensure convergence and prevent overfitting, and finally outputs key prediction values, which is helpful for the operation and resource allocation of the virtual power plant.

[0076] (2) In the present invention, the scheduling optimization module integrates joint prediction, reinforcement learning and multi-objective evolutionary algorithms, which can accurately predict and make intelligent decisions to optimize the scheduling plan of the virtual power plant. It can significantly reduce operating costs, reduce carbon emissions, and improve the utilization efficiency of renewable energy while ensuring the energy balance of the system and the safety of equipment. The module supports dynamic adjustment of scheduling strategies according to different needs, provides flexible low-carbon or cost-priority solutions, and finally generates an optimal scheduling time series that can be directly deployed. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 This is a system block diagram of Embodiment 1 of the present invention;

[0078] Figure 2 Flowchart of the training steps of the joint prediction model in the present invention;

[0079] Figure 3 This is a flow chart of the virtual power plant optimization scheduling method in the present invention. DETAILED DESCRIPTION

[0080] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making creative work shall fall within the scope of protection of the present invention.

[0081] Example 1: Figure 1 and Figure 2 As shown, the present invention proposes a blockchain-based virtual power plant optimization and scheduling system, which is applied to a virtual power plant management platform and includes:

[0082] The data acquisition module is used to collect real-time operating data of the virtual power plant, including photovoltaic device power generation, wind turbine power generation, energy storage device discharge power, photovoltaic panel irradiation intensity, wind turbine wind speed, remaining energy storage device power, load-side power consumption, ambient temperature, grid voltage, and grid frequency. The real-time collected virtual power plant operating data is stored on the blockchain, and the stored data is hashed and recorded on the blockchain to ensure data integrity and immutability. All participants can view the data on the blockchain, improving the transparency of the system. The immutable nature of the blockchain makes it easy to trace historical data, facilitating audits and compliance checks.

[0083] The data acquisition module significantly improves the system's decision-making accuracy and response speed by collecting real-time operating data of the virtual power plant, including the power generation power of photovoltaic devices and wind turbines, the discharge power and remaining power of energy storage equipment, load-side power consumption, environmental parameters (such as temperature, irradiation intensity, wind speed), and grid status (voltage and frequency); these data support accurate prediction and optimized scheduling, help identify inefficient links or potential failures in the system, improve operational efficiency and reduce waste; combined with blockchain technology to ensure data transparency and security, simplify the audit process, and meet regulatory requirements; overall, this module not only enhances the flexibility and stability of the system, but also improves user experience and service quality, effectively improving the overall performance and competitiveness of the virtual power plant.

[0084] The energy load joint prediction module is used to build a joint prediction model based on the historical operation data of the virtual power plant using the LSTM algorithm. The joint prediction model can predict the power generation of each distributed energy source in the virtual power plant and the load demand of the entire virtual power plant in the future.

[0085] The energy load joint forecasting module processes the historical operation data of the virtual power plant in the following steps:

[0086] Acquire historical operating data of the virtual power plant and perform preprocessing operations on the historical operating data of the virtual power plant, the preprocessing operations including but not limited to data cleaning, denoising and normalization;

[0087] Convert the pre-processed virtual power plant operation data into a time series format, treat the data of each time step as a vector, set a time window (for example, 24 hours), and treat the data within the window as an input sequence;

[0088] The time series data is divided into training set, validation set and test set in chronological order, with the allocation ratio of 70% training, 15% validation and 15% test;

[0089] The pre-processed virtual power plant operation data is securely stored on the blockchain, hashed, and recorded on the blockchain.

[0090] The process of building a joint prediction model in the energy load joint prediction module includes:

[0091] Construct a joint prediction model to predict the power generation of each distributed energy source in the virtual power plant and the load demand of the entire virtual power plant over a period of time in the future. The model structure includes the following parts:

[0092] Input layer: Determine the input layer dimension p to be 10, and construct the input vector x by combining the data of each time step, x = [x1, x2, ..., x t …, x m ], where xt ∈R p is the input vector at time t, R p Represents a p-dimensional real number space, taking x as input data;

[0093] LSTM layer: The LTSM layer contains m LTSM units. At each time step t, the input data x will be processed in parallel by m LSTM units. The output dimension of each LSTM unit is d.

[0094] Fully connected layer: receives the output from the LSTM layer Map it to a 4-dimensional output vector Calculated using the following formula:

[0095]

[0096] Where W fc is the weight matrix of the fully connected layer, b fc is the bias vector of the fully connected layer, z is the linear transformation result of the fully connected layer, and f(·) is the activation function. Since the prediction is a continuous value, the linear activation function is selected here (for predictions without range restrictions);

[0097] Output layer: directly outputs the prediction vector The predicted values ​​of photovoltaic power generation, wind turbine power generation, energy storage device discharge, and total load are expressed as follows:

[0098]

[0099] Data is collected in real time from the distributed energy resources, load side, and grid parameter sensors of the virtual power plant, and preprocessed through cleaning, denoising, and normalization. The preprocessed data is uploaded to the blockchain network and stored using blockchain's distributed ledger technology. Each block contains data records for a period of time, and each block is linked to the previous block through a hash value, forming a chain structure to ensure that the data cannot be tampered with.

[0100] Before uploading data to the blockchain, it can be encrypted and digitally signed to ensure confidentiality and authenticity of the data. The distributed nature of the blockchain means that data is not stored in one place, but is dispersed across multiple nodes in the network, improving data security and anti-attack capabilities. The data in each block is hashed and its validity is verified through a consensus mechanism to prevent data tampering.

[0101] The process of the LSTM layer in the joint prediction model processing input data includes:

[0102] For each time step t, the computation process of the m-th LSTM unit is as follows:

[0103] If m=1, the input is x i , otherwise the input is the output of the previous layer (m-1) Calculated by the following formula:

[0104]

[0105] Where σ represents the sigmoid activation function, ⊙ represents element-wise multiplication, are the weight matrices of the input gate, forget gate, cell state, and output gate, respectively. are the bias vectors of the input gate, forget gate, cell state, and output gate respectively;

[0106] The training steps for building the joint prediction model of the energy load joint prediction module are as follows:

[0107] T1: Randomly initialize all weight matrices and bias vectors;

[0108] T2: Input the training data into the joint prediction model to perform calculations on the input layer, LSTM layer, fully connected layer, and output layer to obtain the prediction vector

[0109] T3: Use mean square error as the loss function: Where N is the number of training samples, y t is a real-valued vector;

[0110] T4: Use the backpropagation algorithm to calculate the gradient of the loss function with respect to all model parameters. This requires backpropagating the error from the output layer to the multi-layer LSTM layer and calculating the gradient of each LSTM unit. Using the backpropagation algorithm to calculate the gradient of the loss function with respect to all model parameters is an existing technology and will not be elaborated on here.

[0111] T5: Use an optimizer (such as Adam and RMSprop) to update the model parameters and adjust the weight matrix and bias vector based on the calculated gradient. The learning rate is an important hyperparameter and needs to be adjusted. Updating model parameters and adjusting the learning rate using the optimizer are both existing technologies and will not be elaborated on here.

[0112] T6: Repeat steps T2-T5 until the model converges or reaches the preset stopping condition (for example, the loss on the validation set no longer decreases). Early stopping can be used to prevent overfitting.

[0113] After model training is completed, the performance of the trained model is evaluated using the test set. Using the test set to evaluate model performance is an existing technology and will not be elaborated on here.

[0114] The energy load joint prediction module uses the LSTM algorithm to build a model, achieving high-precision prediction of the power generation of each distributed energy in the virtual power plant and the load demand of the entire virtual power plant in the future period; the module first ensures the quality of the input data by performing detailed data cleaning, denoising and normalization preprocessing on historical operation data, and then uses the LSTM network's ability to capture long-term dependencies in time series data to effectively improve the prediction accuracy; in addition, its design is flexible and can adjust the input feature dimension and the number of LSTM layer units according to different application scenarios, enhancing adaptability and scalability; at the same time, by converting historical data into time series form and using the sliding window method to organize training samples, the system's real-time learning and prediction capabilities are supported; during the training process, random initialization parameters, division of training set / validation set / test set, mean square error loss function, backpropagation algorithm and optimizer (such as Adam and RMSprop) are adopted to ensure effective training and rapid convergence of the model, and the early stopping method is used to prevent overfitting, further improving the model's generalization ability. The final output directly provides demand forecast values ​​for photovoltaic device power generation, wind turbine power generation, energy storage device discharge and total load, which is crucial for the operation and management of virtual power plants and the optimal allocation of resources, reflecting its strong practicality.

[0115] Embodiment 2: The technical solution of this embodiment of the present invention differs from that of Embodiment 1 in that:

[0116] like Figure 1 As shown in the figure, the scheduling optimization module is used to introduce reinforcement learning and multi-objective evolutionary algorithm to build an intelligent scheduling optimization model based on the prediction results of the joint prediction model. The model is used to optimize the distributed energy generation parameters and generate the optimal scheduling plan for the virtual power plant operation;

[0117] The process of introducing reinforcement learning and multi-objective evolutionary algorithms into the scheduling optimization module to build an intelligent scheduling optimization model includes:

[0118] Get the prediction result of the joint prediction model. The prediction result of the joint prediction model is the prediction vector

[0119] The distributed energy generation parameters are defined as vector r = [P gf , P fl , P cl ] T , where P gf is the power generated by the photovoltaic device, P fl is the power generated by the wind turbine, P cl Discharge power for energy storage equipment;

[0120] The model is built through the following multi-objective function, with the goals of minimizing the total comprehensive operating cost, minimizing carbon emissions and maximizing the utilization of renewable energy;

[0121] The virtual power plant cost function is as follows:

[0122] Among them C zy is the total operating cost, C gm is the unit price of electricity purchased from the power grid, C cj is the energy storage charging price, P gm,t is the amount of electricity purchased from the grid, which represents the power purchased from the grid at time step t, is the energy storage charging power, which represents the power of the energy storage device charged from the grid at time step t;

[0123] The carbon emission function of a virtual power plant is as follows:

[0124] Among them E zy is the total carbon emissions, γ is the carbon emission coefficient;

[0125] The renewable energy utilization rate function of the virtual power plant is as follows:

[0126] Among them U nl is the utilization rate of renewable energy;

[0127] The energy balance constraint is as follows: P gf,t +P fl,t +P cl,t +P gm,t =D xz,t ;

[0128] The power generation constraints are as follows:

[0129] The operating constraints of energy storage equipment are as follows:

[0130]

[0131]

[0132] SOC min ≤SOC t ≤SOC max ;

[0133] Among them, SOC t Represents the remaining power of the energy storage device at time step t, η c Represents the charging efficiency of energy storage equipment, η d Indicates the discharge efficiency of the energy storage device, Represents the discharge power of the energy storage device, P cl,maxIndicates the maximum output power of the energy storage device, SOC min Indicates the minimum power of the energy storage device, SOC max Indicates the maximum capacity of the energy storage device;

[0134] Designing a reinforcement learning framework:

[0135] State space S t for:

[0136] Action space A t For: A t =[P gf,t ,P fl,t ,P cl,t ];

[0137] Reward function R t For: R t =-(αC zy,t +βE zy,t -λU nl,t ), where α, β, and λ are weight coefficients used to balance economic efficiency, environmental protection, and renewable utilization rate;

[0138] The process by which the scheduling optimization module optimizes distributed energy generation parameters and generates the optimal scheduling plan for virtual power plant operation includes:

[0139] The multi-objective evolutionary algorithm solves the Pareto optimal scheduling strategy as follows: the action sequence generated by reinforcement learning is used as the initial population and input into the multi-objective evolutionary algorithm for further optimization. The specific steps are as follows:

[0140] R1: Encoding scheduling scheme is chromosome

[0141] R2: Construct the objective function vector: F(x) = [C zy (d),E zy (d),-U nl (d)];

[0142] R3: Generate new scheduling strategy combinations using selection, crossover, and mutation operations;

[0143] R4: Select the non-inferior frontier solution set by Pareto superiority sorting

[0144] According to the scheduling priority, the solution set is selected from the non-inferior frontier Select the optimal scheduling time series solution:

[0145] The optimal scheduling plan generated by the scheduling optimization module is stored on the blockchain to ensure data security and immutability. The data in each block is hashed and verified for validity through a consensus mechanism to prevent data tampering.

[0146] This scheduling optimization module integrates reinforcement learning and a multi-objective evolutionary algorithm to construct an intelligent model to optimize distributed energy parameters and generate the optimal scheduling solution for the virtual power plant. Its advantage lies in its comprehensive consideration of operating costs, carbon emissions, and renewable energy utilization, balancing economic efficiency, environmental protection, and resource utilization efficiency. Reinforcement learning improves the model's environmental adaptability, while the multi-objective evolutionary algorithm enhances its solution space search capabilities, collaboratively achieving robust optimization results. The module fully considers multiple constraints such as energy balance, power generation, and energy storage operation, demonstrating excellent flexibility and adaptability. By screening non-inferior solutions through the Pareto front and selecting the optimal scheduling strategy based on actual needs, the scientific and rational nature of decision-making is enhanced. A scheduling mechanism based on predictive information enhances the system's foresight and overall improves the greenness and operational efficiency of the virtual power plant.

[0147] The scheduling execution module is used to control the corresponding execution agencies to execute control measures according to the generated operation scheduling plan;

[0148] The process in which the scheduling execution module controls the corresponding execution agencies to execute control measures according to the generated operation scheduling plan includes:

[0149] After receiving the operation scheduling plan, the scheduling execution module immediately sends a control instruction to the corresponding execution agency. The execution agency executes the corresponding control measures according to the control instruction. The specific contents of the control measures are as follows:

[0150] Adjust the output of the photovoltaic inverter to adjust the power generated by the photovoltaic device from the current value to the optimal value;

[0151] By adjusting the angle or speed of the wind turbine blades, the wind turbine generator power can be adjusted to the optimal value;

[0152] Control the charging and discharging process of the energy storage device to achieve the optimal discharge power while ensuring that the battery health is not affected;

[0153] When necessary, virtual power plants can support the frequency and voltage stability of the grid by adjusting the output of internal power generation units;

[0154] For controllable loads, such as large electrical equipment in industrial users or air conditioning systems in commercial buildings, adjustments can be made according to the scheduling plan to balance supply and demand;

[0155] The dispatching execution module precisely controls photovoltaic inverters, wind turbines, and energy storage equipment to their optimal state based on the operation dispatch plan, dynamically responds to changes in power demand, and ensures the stability of grid frequency and voltage. During the optimization process, the module pays special attention to battery health to avoid damage to ensure long-term stable operation, and finely balances supply and demand by managing adjustable loads (such as industrial high-power equipment and commercial building air-conditioning systems), reducing peak-to-valley differences and improving grid efficiency. In addition, the virtual power plant uses internal power generation units to flexibly support grid stability, provide rapid response to grid fluctuations, and enhance grid reliability and resilience. Overall, the dispatching execution module improves the efficiency of power resource utilization, reduces operating costs, promotes the integration and utilization of clean energy, and provides key support for the efficient management and sustainable development of smart grids.

[0156] Embodiment 3: The technical solution of the embodiment of the present invention is different from that of the embodiment 1 and the embodiment 2 in that:

[0157] like Figure 3 As shown, a blockchain-based virtual power plant optimization scheduling method includes the following steps:

[0158] Step 1: Real-time collection of virtual power plant operating data, including photovoltaic device power generation, wind turbine power generation, energy storage device discharge power, photovoltaic panel irradiation intensity, wind turbine wind speed, energy storage device remaining power, load-side power consumption, ambient temperature, grid voltage, and grid frequency;

[0159] Step 2: Based on the historical operating data of the virtual power plant, a joint prediction model is constructed using the LSTM algorithm. This joint prediction model can predict the power generation of each distributed energy source in the virtual power plant and the load demand of the entire virtual power plant over a period of time in the future.

[0160] Step 3: Based on the prediction results of the joint prediction model, reinforcement learning and multi-objective evolutionary algorithms are introduced to build an intelligent scheduling optimization model. The model is used to optimize the distributed energy generation parameters and generate the optimal scheduling plan for the virtual power plant operation.

[0161] Step 4: Control the corresponding executive agencies to execute control measures according to the generated operation scheduling plan;

[0162] This blockchain-based virtual power plant optimization and scheduling method integrates LSTM prediction, reinforcement learning and multi-objective evolutionary algorithms to achieve accurate prediction and intelligent scheduling optimization of distributed energy generation and load demand, improving energy utilization efficiency and system response speed; blockchain technology ensures data security and transparency, promotes multi-party collaboration, enhances the flexibility, stability and scalability of the system, and provides strong support for the efficient, safe and sustainable operation of virtual power plants.

[0163] The above are only preferred specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solutions and improved concepts of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A blockchain-based virtual power plant optimization and scheduling system, applied to a virtual power plant management platform, characterized in that: include: The data acquisition module is used to collect real-time operating data of the virtual power plant, including photovoltaic device power generation, wind turbine power generation, energy storage device discharge power, photovoltaic panel irradiation intensity, wind turbine wind speed, energy storage device remaining power, load-side power consumption, ambient temperature, grid voltage, and grid frequency; The energy load joint prediction module is used to build a joint prediction model based on the historical operation data of the virtual power plant using the LSTM algorithm. The joint prediction model can predict the power generation of each distributed energy source in the virtual power plant and the load demand of the entire virtual power plant in the future. The scheduling optimization module is used to build an intelligent scheduling optimization model based on the prediction results of the joint prediction model by introducing reinforcement learning and multi-objective evolutionary algorithms. The model is used to optimize the distributed energy generation parameters and generate the optimal scheduling plan for the virtual power plant operation. The scheduling execution module is used to control the corresponding execution agencies to execute control measures according to the generated operation scheduling plan.

2. A blockchain-based virtual power plant optimization scheduling system according to claim 1, characterized in that: The process of the energy load joint forecasting module processing the historical operation data of the virtual power plant includes: Obtain historical operating data of the virtual power plant and perform preprocessing operations on the historical operating data of the virtual power plant; Convert the pre-processed virtual power plant operation data into a time series form, treat the data of each time step as a vector, set a time window, and treat the data within the window as an input sequence; The time series data is divided into training set, validation set and test set in chronological order, with the allocation ratio of 70% training, 15% validation and 15% test.

3. The blockchain-based virtual power plant optimization scheduling system according to claim 1 is characterized in that: The process of constructing the joint prediction model by the energy load joint prediction module includes: Construct a joint prediction model to predict the power generation of each distributed energy source in the virtual power plant and the load demand of the entire virtual power plant over a period of time in the future. The model structure includes the following parts: Input layer: Determine the input layer dimension p to be 10, and construct the input vector x by combining the data of each time step, x = [x1, x2, ..., x t …, x m ], where x t ∈R p is the input vector at time t, R p Represents a p-dimensional real number space, taking x as input data; LSTM layer: The LTSM layer contains m LTSM units. At each time step t, the input data x will be processed in parallel by m LSTM units. The output dimension of each LSTM unit is d. Fully connected layer: receives the output from the LSTM layer Map it to a 4-dimensional output vector Calculated using the following formula: Where W fc is the weight matrix of the fully connected layer, b fc is the bias vector of the fully connected layer, z is the linear transformation result of the fully connected layer, and f(·) is the activation function; Output layer: directly outputs the prediction vector The predicted values ​​of photovoltaic power generation, wind turbine power generation, energy storage device discharge, and total load are expressed as follows:

4. A blockchain-based virtual power plant optimization scheduling system according to claim 3, characterized in that: The process of the LSTM layer in the joint prediction model processing input data includes: For each time step t, the computation process of the m-th LSTM unit is as follows: If m=1, the input is x i , otherwise the input is the output of the previous layer (m-1) Calculated by the following formula: Where σ represents the sigmoid activation function, ⊙ represents element-wise multiplication, are the weight matrices of the input gate, forget gate, cell state, and output gate, respectively. are the bias vectors of the input gate, forget gate, cell state, and output gate, respectively.

5. The blockchain-based virtual power plant optimization scheduling system according to claim 4 is characterized in that: The training steps for constructing the joint prediction model of the energy load joint prediction module are as follows: T1: Randomly initialize all weight matrices and bias vectors; T2: Input the training data into the joint prediction model to perform calculations on the input layer, LSTM layer, fully connected layer, and output layer to obtain the prediction vector T3: Use mean square error as the loss function: Where N is the number of training samples, y t is a real-valued vector; T4: Use the backpropagation algorithm to calculate the gradient of the loss function with respect to all model parameters; T5: Use the optimizer to update the model parameters and adjust the weight matrix and bias vector according to the calculated gradient; T6: Repeat steps T2-T5 until the model converges or the preset stopping condition is reached.

6. The blockchain-based virtual power plant optimization scheduling system according to claim 1 is characterized in that: The process of introducing reinforcement learning and multi-objective evolutionary algorithm into the scheduling optimization module to build an intelligent scheduling optimization model includes: Get the prediction result of the joint prediction model. The prediction result of the joint prediction model is the prediction vector The distributed energy generation parameters are defined as vector r = [P gf , P fl , P cl ] T , where P gf is the power generated by the photovoltaic device, P fl is the power generated by the wind turbine, P cl Discharge power for energy storage equipment; The model is built through the following multi-objective function, with the goals of minimizing the total comprehensive operating cost, minimizing carbon emissions and maximizing the utilization of renewable energy; The virtual power plant cost function is as follows: Among them C zy is the total operating cost, C gm is the unit price of electricity purchased from the power grid, C cj is the energy storage charging price, P gm,t is the amount of electricity purchased from the grid, which represents the power purchased from the grid at time step t, is the energy storage charging power, which represents the power of the energy storage device charged from the grid at time step t; The carbon emission function of a virtual power plant is as follows: Among them E zy is the total carbon emissions, γ is the carbon emission coefficient; The renewable energy utilization rate function of the virtual power plant is as follows: Among them U nl is the utilization rate of renewable energy; The energy balance constraint is as follows: P gf,t +P fl,t +P cl,t +P gm,t =D xz,t ; The power generation constraints are as follows: The operating constraints of energy storage equipment are as follows: SOC min ≤SOC t ≤SOC max ; Among them, SOC t Represents the remaining power of the energy storage device at time step t, η c Represents the charging efficiency of energy storage equipment, η d Indicates the discharge efficiency of the energy storage device, Represents the discharge power of the energy storage device, P cl,max Indicates the maximum output power of the energy storage device, SOC min Indicates the minimum power of the energy storage device, SOC max Indicates the maximum capacity of the energy storage device; Designing a reinforcement learning framework: State space S t for: Action space A t For: A t =[P gf,t ,P fl,t ,P cl,t ]; Reward function R t For: R t =-(αC zy,t +βE zy,t -λU nl,t ), where α, β, and λ are weight coefficients.

7. A blockchain-based virtual power plant optimization scheduling system according to claim 6, characterized in that: The process of the scheduling optimization module optimizing the distributed energy generation parameters and generating the optimal scheduling plan for the virtual power plant operation includes: The multi-objective evolutionary algorithm solves the Pareto optimal scheduling strategy as follows: the action sequence generated by reinforcement learning is used as the initial population and input into the multi-objective evolutionary algorithm for further optimization. The specific steps are as follows: R1: Encoding scheduling scheme is chromosome R2: Construct the objective function vector: F(x) = [C zy (d),E zy (d),-U nl (d)]; R3: Generate new scheduling strategy combinations using selection, crossover, and mutation operations; R4: Select the non-inferior frontier solution set by Pareto superiority sorting According to the scheduling priority, the solution set is selected from the non-inferior frontier Select the optimal scheduling time series solution:

8. The blockchain-based virtual power plant optimization scheduling system according to claim 1 is characterized in that: The process of the scheduling execution module controlling the corresponding execution mechanism to execute the control measures according to the generated operation scheduling plan includes: After receiving the operation scheduling plan, the scheduling execution module immediately sends a control instruction to the corresponding execution agency. The execution agency executes the corresponding control measures according to the control instruction. The specific contents of the control measures are as follows: Adjust the output of the photovoltaic inverter to adjust the power generated by the photovoltaic device from the current value to the optimal value; By adjusting the angle or speed of the wind turbine blades, the wind turbine generator power can be adjusted to the optimal value; Control the charging and discharging process of the energy storage device so that the discharge power of the energy storage device reaches the optimal value while ensuring that the battery health status is not affected.

9. A virtual power plant optimization scheduling method based on blockchain, characterized in that: The following steps are involved: Step 1: Real-time collection of virtual power plant operating data, including photovoltaic device power generation, wind turbine power generation, energy storage device discharge power, photovoltaic panel irradiation intensity, wind turbine wind speed, energy storage device remaining power, load-side power consumption, ambient temperature, grid voltage, and grid frequency; Step 2: Based on the historical operating data of the virtual power plant, a joint prediction model is constructed using the LSTM algorithm. This joint prediction model can predict the power generation of each distributed energy source in the virtual power plant and the load demand of the entire virtual power plant over a period of time in the future. Step 3: Based on the prediction results of the joint prediction model, reinforcement learning and multi-objective evolutionary algorithms are introduced to build an intelligent scheduling optimization model. The model is used to optimize the distributed energy generation parameters and generate the optimal scheduling plan for the virtual power plant operation. Step 4: Control the corresponding executive agencies to execute control measures according to the generated operation scheduling plan.

Images