Virtual power plant intelligent matching and optimal scheduling method and system based on artificial intelligence, and computer program product

Abstract

The invention discloses a virtual power plant intelligent matching and optimal scheduling method and system based on artificial intelligence and a computer program product, and belongs to the technical field of virtual power plant intelligent matching and optimal scheduling. Establishing a virtual power plant intelligent matching model based on a deep Q network method, and training to obtain a value function; judging whether the value function converges or not, and if yes, obtaining an optimal virtual power plant resource intelligent matching combination; and if not, sending the virtual power plant resource selection combination obtained in the training process to the optimal power flow calculation model, and returning the obtained line load power to the virtual power plant intelligent matching model for optimization. According to the invention, the operation safety and economy of the power distribution network can be improved, the new energy consumption capability is improved, the adjustment flexibility of the power distribution network is enhanced, and an intelligent solution is provided for efficient utilization of large-scale distributed energy.

Description

Technical Field

[0001] This application relates to the field of intelligent matching and optimal scheduling of virtual power plants, and particularly to an intelligent matching and optimal scheduling method, system, and computer program product of a virtual power plant based on artificial intelligence. Background Art

[0002] With the large-scale access of new energy to the distribution network, virtual power plants, as a key technology for aggregating distributed resources, play an important role in alleviating line overloads and optimizing grid operation. Currently, the scheduling of virtual power plants for distribution network overload problems mainly uses traditional optimization methods such as linear programming, integer programming, genetic algorithms, and particle swarm optimization. These methods usually rely on deterministic mathematical models, assuming that the output of new energy and load demand can be accurately predicted. However, in actual operation, the output of wind and light has strong volatility, and the load also has significant uncertainties, resulting in difficulties for traditional methods to dynamically adjust scheduling strategies and poor adaptability. In addition, with the expansion of the scale of the distribution network, the quantity and variety of virtual power plant resources have increased sharply. The computational complexity of traditional optimization methods in high-dimensional state spaces increases exponentially, the solution efficiency is low, it is difficult to meet the requirements of real-time optimal scheduling, and the optimization process highly depends on manual experience adjustment, lacking the ability of autonomous learning and environmental interaction, and unable to achieve adaptive optimization under complex and changeable grid operation conditions.

[0003] On the other hand, current methods mostly focus on the maximum output capacity of virtual power plants, while ignoring key indicators such as response latency, output stability, and line load matching degree, resulting in deviations between scheduling schemes and actual demands, affecting the regulation effect and economy of the distribution network.

[0004] To overcome the above defects, there is an urgent need for an intelligent scheduling method that can adapt to new energy fluctuations and load changes and take into account multi-objective optimization. Summary of the Invention

[0005] This application aims to provide an intelligent matching and optimal scheduling method, system, and computer program product of a virtual power plant based on artificial intelligence, to solve the deficiencies of the prior art in dealing with new energy volatility, load uncertainty, and resource matching optimization, achieve precise and efficient virtual power plant resource scheduling, reduce the overload rate of the distribution network, and improve the economy and reliability of system operation.

[0006] To achieve the above object, the technical solution of this application is: An intelligent matching and optimal scheduling method of a virtual power plant based on artificial intelligence, including, Step S1: Taking the minimum expected value of the power shortage of all new energy in the distribution network as the objective function, constructing an aggregation analysis model of the virtual power plant considering new energy, and calculating all virtual power plant resources in the distribution network through the particle swarm optimization algorithm; Step S2: Establish measurement indexes for virtual power plant resources based on all virtual power plant resources in the distribution network; Step S3: Based on the established measurement indexes for virtual power plant resources, establish an intelligent matching model for virtual power plants using the deep Q-network method, and train the intelligent matching model for virtual power plants to obtain a value function; Step S4: Determine whether the value function converges. If it does, proceed to Step S6; if not, proceed to Step S5; Step S5: Send the virtual power plant resource selection combination obtained during the training process to the optimal power flow calculation model. The optimal power flow calculation model aims to minimize the price of the virtual power plant resources used, and returns the line load power obtained from the optimal power flow calculation of the distribution network after using the virtual power plant resource selection combination to the intelligent matching model for virtual power plants, obtain the state space after the distribution network environment is updated, and perform training again. Return to Step S4; Step S6: Obtain the optimal intelligent matching combination of virtual power plant resources.

[0007] Optionally, the objective function of the virtual power plant aggregation analysis model is expressed as follows: Among them, represents the objective function of minimizing the expected value of the power shortage of all new energy sources in the distribution network, E represents the expected value operator, D represents the load demand of the distribution network, represents the total output of all new energy sources in the distribution network; The constraint conditions of the virtual power plant aggregation analysis model include the output constraints of photovoltaic resources and wind power resources, which are expressed as follows: Among them, represents the i-th photovoltaic resource, represents the j-th wind power resource, N, M respectively represent the quantities of photovoltaic resources and wind power resources; 、 respectively represent the upper and lower limits of the output of photovoltaic resources, 、 respectively represent the upper and lower limits of the output of wind power resources.

[0008] Optionally, the measurement indexes for virtual power plant resources include: line load and the response latency of virtual power plant resources, the coincidence degree of the output of virtual power plant resources and line load during heavy load periods, and the average value of the output of virtual power plant resources during heavy load periods; The line load and the response latency of virtual power plant resources are expressed as follows: Among them, S1 is the response latency of line load and virtual power plant resources, indicating the i overall correlation coefficient between the line load of the j th line and the virtual power plant resources of the th virtual power plant resource. i represents the line load of the th line, and represents the jth virtual power plant resource; Among them, S 2 is the second virtual power plant resource measurement index, represents the heavy load period of the i th line, represents the value corresponding to the line load of the ith line during the heavy load period, represents the value corresponding to the jth virtual power plant resource during the heavy load period, represents the weight used to calculate the output coincidence degree; The average value of the output of virtual power plant resources during the heavy load period is expressed as follows: Among them, S 3 is the third virtual power plant resource measurement index, represents the heavy load period of the ith line.

[0009] Optionally, step S3 includes: Obtain the state of virtual power plant resources at the current moment from the virtual power plant resource environment and extract the required data; Establish the state space of the deep Q-network method according to the virtual power plant resource measurement index; Establish the action space of the deep Q-network method; Design the reward function of the deep Q-network method according to the virtual power plant resource measurement index; Train the established deep Q-network model to obtain the corresponding action-value function.

[0010] Optionally, the required data extracted includes: line load power, line rated power, and line heavy load period power; The corresponding action-value function is expressed as follows: Among them, represents the weight matrix of the virtual power plant intelligent matching model, represents the value obtained by making an action selection based on the action-value function; represents the state-action space at time t,A t represents Y t the corresponding action space, represents the action space at the next moment, represents the state at the next moment.

[0011] Optionally, the judgment formula for the convergence of the trained value function is expressed as follows: where, represents Q the rate of change of the value, T represents the number of samples in the experience pool, represents the k weight matrix of the th training cycle, k-1 represents the weight matrix of the th training cycle; when the training converges, otherwise, it does not converge;

[0012] Optionally, the objective function of the optimal power flow calculation model, the first constraint condition of the optimal power flow calculation model, and the second constraint condition of the optimal power flow calculation model are expressed as follows: where, represents the objective function of the optimal power flow calculation model, const 1.1 represents the first constraint condition of the optimal power flow calculation model, const 1.2 represents the second constraint condition of the optimal power flow calculation model; a , b are the node numbers of the low-voltage distribution network; s is the node number of the low-voltage distribution network connected to the medium-voltage distribution network, and the subscript contains s indicating that the parameter belongs to the low-voltage distribution network parameter of the medium-voltage distribution network node s ; , are the a , b voltage values of node t at the time period; t is the ab active power flowing through the head end of the branch at the , are the ab impedances of the branch; is the t reactive power flowing through the head end of the branch at the ab time period; represents t the branch at theab The current flowing through; and are respectively t time period nodes b the active power and reactive power injected; and are respectively t the active power and reactive power at the head end of the branch at a certain moment bk ; k: b → k is expressed as the set of all child nodes with b the node as the parent node; and are respectively the upper limit value and the lower limit value of the node voltage of the low - voltage distribution network; represents the set of the selected combination of virtual power plant resources used, and represents the price of the i - th virtual power plant resource.

[0013] Optionally, the optimal intelligent matching combination of virtual power plant resources is represented as follows: Among them, represents the optimal virtual power plant resource selection action, and represents the optimal state value function.

[0014] An artificial - intelligence - based virtual power plant intelligent matching and optimal scheduling system for performing the artificial - intelligence - based virtual power plant intelligent matching and optimal scheduling method as described in any one of the above, including: a virtual power plant resource calculation module, a virtual power plant resource measurement index establishment module, a virtual power plant resource intelligent matching module, and a virtual power plant combination optimization module; The virtual power plant resource calculation module, the virtual power plant resource measurement index establishment module, the virtual power plant resource intelligent matching module, and the virtual power plant combination optimization module are connected in sequence.

[0015] A computer program product, including a computer program, which implements the steps of the artificial - intelligence - based virtual power plant intelligent matching and optimal scheduling method as described in any one of the above when executed by a processor.

[0016] The virtual power plant intelligent matching and optimal scheduling method, system and computer program product based on artificial intelligence provided by this application construct a virtual power plant intelligent matching model through the deep Q-network method to address the deficiencies of existing technologies in dealing with new energy volatility, load uncertainty, and resource matching optimization, and have significant advantages in dynamic adaptability, accuracy, and economy. Through the self-learning ability of reinforcement learning, the system can autonomously optimize the scheduling strategy, dynamically adapt to the fluctuations in new energy output and load changes, and overcome the defects of relying on manual experience and being difficult to cope with uncertainties. At the same time, this application not only considers the maximum output capacity of virtual power plant resources, but also comprehensively introduces measurement indicators such as line load matching degree and response delay to make resource scheduling more accurate, thereby effectively reducing the line overload rate and improving the safety and stability of the power grid. In addition, combined with the optimal power flow calculation model, with the goal of minimizing the usage cost of the virtual power plant, while ensuring the safe operation of the power grid, the economy is optimized, providing a more efficient and flexible solution for new energy consumption and intelligent power grid scheduling.

[0017] This application can not only improve the safety and economy of the distribution network operation, but also enhance the new energy consumption capacity, increase the flexibility of the power grid regulation, and provide an intelligent solution for the efficient utilization of large-scale distributed energy.

[0018] To make the above features and advantages of the application more obvious and understandable, specific embodiments are given below and detailed descriptions are made in conjunction with the accompanying drawings as follows. Description of the Drawings

[0019] Figure 1 It is a flowchart of the virtual power plant intelligent matching and optimal scheduling method based on artificial intelligence provided by this application.

[0020] Figure 2 It is a structural diagram of the deep Q-network method adopted by this application.

[0021] Figure 3 It is a module diagram of the virtual power plant intelligent matching and optimal scheduling system based on artificial intelligence provided by this application. Detailed Embodiments

[0022] To make the objectives and technical solutions of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of this application. Obviously, the described embodiments are some but not all of the embodiments of this application. All other embodiments obtained by those of ordinary skill in the art based on the described embodiments of this application without creative efforts shall fall within the scope of protection of this application.

[0023] In a specific embodiment of this application, please refer to Figure 1 , Figure 1This is the flowchart of the intelligent matching and optimal scheduling method for a virtual power plant based on artificial intelligence provided by this application. The intelligent matching and optimal scheduling method for a virtual power plant based on artificial intelligence provided by this application includes: steps S1 to S6.

[0024] Step S1: Taking the minimum expected value of the power shortage of all new energies in the distribution network as the objective function, construct an aggregated analysis model of the virtual power plant considering new energies, and calculate all virtual power plant resources in the distribution network through the particle swarm optimization algorithm (PSO); Step S2: Based on all virtual power plant resources in the distribution network, establish a measurement index for virtual power plant resources; Step S3: Based on the established measurement index for virtual power plant resources, establish an intelligent matching model for the virtual power plant using the deep Q-network (DQN) method, and train the intelligent matching model for the virtual power plant to obtain a value function; Step S4: Determine whether the value function converges. If it does, enter step S6; if not, enter step S5; Step S5: Send the virtual power plant resource selection combination obtained during the training process to the optimal power flow calculation model. The optimal power flow calculation model aims to minimize the price of the virtual power plant resources used, and returns the line load power obtained by calculating the optimal power flow of the distribution network after using the virtual power plant resource selection combination to the intelligent matching model of the virtual power plant to obtain the state space updated by the distribution network environment Train again and return to step S4; Step S6: Obtain the optimal intelligent matching combination of virtual power plant resources.

[0025] In step S1, please refer to Figure 1 step S1 in, taking the minimum expected value of the power shortage of all new energies in the distribution network as the objective function, construct an aggregated analysis model of the virtual power plant considering new energies, and calculate all virtual power plant resources in the distribution network through the particle swarm optimization algorithm.

[0026] As an example, the objective function of the minimum expected value of the power shortage of all new energies in the distribution network is expressed as follows: (1) Among them, represents the objective function of the minimum expected value of the power shortage of all new energies in the distribution network, E represents the expected value operator, D represents the load demand of the distribution network, represents the total output of all new energies in the distribution network, and its acquisition process is as follows: First, obtain the historical data of all new energies owned by the distribution network, clean and preprocess all the data. Assume that all the photovoltaic resources contained in the distribution network are , and all the wind power resources are The total output of all new energy sources in this distribution network P total is calculated as follows: (2) Among them, represents the total output of new energy sources contained in the distribution network, represents the i-th photovoltaic resource, represents the j-th wind power resource, N, M respectively represent the quantities of photovoltaic resources and wind power resources.

[0027] As an example, the constraint conditions of the virtual power plant aggregation analysis model include the output constraints of photovoltaic resources and wind power resources, which are expressed as follows: (3) Among them, , respectively represent the upper and lower limits of the output of photovoltaic resources, , respectively represent the upper and lower limits of the output of wind power resources.

[0028] Furthermore, through the virtual power plant aggregation analysis model based on the objective function and constraint conditions, the particle swarm optimization algorithm is used for solution. Among them, the particle position vector of the particle swarm optimization algorithm represents the output values of each photovoltaic resource and wind power resource, and the particle position vector of the particle swarm optimization algorithm is expressed as follows: (4) Among them, x represents the particle position vector of the particle swarm optimization algorithm, and the fitness function f (x) represents the expected value of the electricity quantity shortage of new energy power, which is expressed as follows: (5) Furthermore, by setting a penalty function through the constraint conditions and performing optimization calculations using the particle swarm optimization algorithm, the optimal output value of each resource is obtained, that is, all virtual power plant resources in the distribution network P VPP value.

[0029] In step S2, please refer to Figure 1 step S2 in, and establish a measurement index for virtual power plant resources according to all virtual power plant resources in the distribution network.

[0030] As an example, the virtual power plant resource measurement index is used to measure the matching degree between the virtual power plant and the line power. The virtual power plant resource measurement index includes: line load and the response latency of virtual power plant resources, the coincidence degree of the output of virtual power plant resources and the line load during heavy load periods, and the average value of the output of virtual power plant resources during heavy load periods.

[0031] Specifically, establishing the virtual power plant resource measurement index includes: steps S21 to S22.

[0032] Step S21: Calculate the heavy load period; Step S22: Construct the virtual power plant resource measurement index based on the values of the line load and virtual power plant resources, which are respectively the response latency of the line load and virtual power plant resources, the coincidence degree of the output of virtual power plant resources and the line load during heavy load periods, and the average value of the output of virtual power plant resources during heavy load periods.

[0033] As an example, in step S21, the heavy load period is expressed as follows: (6) Wherein, represents the heavy load period of the i th line, and respectively represent the start time and end time of the line power curve, represents the rated power of the line, represents the value corresponding to the line load during the heavy load period.

[0034] As an example, in step S22, in order to construct the virtual power plant resource pool, three virtual power plant resource measurement indexes are established. The first virtual power plant resource measurement index is the response latency between the line load and virtual power plant resources, which represents the overall correlation between the line load value and virtual power plant resources. The correlation coefficient between the output curves of all virtual power plant resources on this line and the load curve of the line represents the first virtual power plant resource measurement index, which is expressed as follows: (7) Wherein, S1 is the first virtual power plant resource measurement index, representing the overall correlation coefficient between the i th line load and the j th virtual power plant resource, represents the i th line load, represents the jth virtual power plant resource.

[0035] As an example, the second virtual power plant resource measurement index required for the construction of the virtual power plant resource pool is the coincidence degree of the output of the virtual power plant resources and the line load during the heavy load period, which represents the matching degree between the two during the heavy load period. It is achieved by calculating the coincidence degree of the line load and the output of the virtual power plant during the heavy load period, and is expressed as follows: (8) Among them, S 2 is the second virtual power plant resource measurement index, represents the heavy load period of the i th line, represents the value corresponding to the load of the ith line during the heavy load period, represents the value corresponding to the jth virtual power plant resource during the heavy load period, represents the weight used to calculate the coincidence degree of the output.

[0036] As an example, the third virtual power plant resource measurement index required for the construction of the virtual power plant resource pool is the average value of the output of the virtual power plant resources during the heavy load period, which is expressed as follows: (9) Among them, S 3 is the third virtual power plant resource measurement index, represents the heavy load period of the i th line.

[0037] As an example, through formulas (7), (8), and (9), three virtual power plant resource measurement indexes required for establishing the virtual power plant intelligent matching model in step S3 can be obtained, which can quantify the adjustment ability of the virtual power plant resources to the line in terms of time level and magnitude level.

[0038] In step S3, please refer to Figure 1 for step S3. Based on the established virtual power plant resource measurement indexes, establish a virtual power plant intelligent matching model based on the deep Q-network method, and train the virtual power plant intelligent matching model to obtain the value function.

[0039] As an example, step S3 includes: steps S31 to S36.

[0040] Step S31: Obtain the virtual power plant resource status at the current moment from the virtual power plant resource environment and extract the required data; Step S32: Establish the state space of the deep Q-network method according to the virtual power plant resource measurement indexes; Step S33: Establish the action space of the deep Q-network method; Step S34: Design the reward function of the deep Q-network method according to the virtual power plant resource measurement indexes; Step S35: Train the established deep Q - network model to obtain the corresponding action - value function.

[0041] As an example, in step S31, the required data extracted includes: line load power, line rated power, and line heavy - load period power.

[0042] As an example, in step S32, establish the state space of the deep Q - network method according to the virtual - power - plant resource measurement index, which is expressed as follows: (10) Among them, LUR represents the line load rate, which is the ratio of the line load power to the line rated power and is used to measure the degree of line overload; P VPP represents all virtual - power - plant resources; S 1 、S 2 、S 3 respectively represent three virtual - power - plant resource measurement indexes; t represents the current moment and is used to capture the load change trend. Among them, the line load rate LUR is calculated as follows: (11) Among them, represents the line rated power, represents the line load power.

[0043] As an example, in step S33, this application defines the action space as discrete actions, that is, which virtual - power - plant resources are selected, representing the virtual - power - plant resource selection combination. The action space is expressed as follows: (12) Among them, N + M is the number of virtual - power - plant resources, where 1 for each resource represents selected and 0 represents not selected; represents the action space.

[0044] As an example, in step S34, design the reward function according to the three virtual - power - plant resource measurement indexes. The reward function includes the response - delay reward and the virtual - power - plant resource matching - degree reward. Among them, the response - delay reward indicates that the higher the overall correlation between the virtual - power - plant resources and the line power, the higher the reward. The response - delay reward is expressed as follows: (13) Among them, represents the response - delay reward, It represents 1 or 0, where 1 indicates the selection of virtual power plant resources and 0 indicates the non - selection of virtual power plant resources.

[0045] Furthermore, the virtual power plant resource matching degree reward means that the closer the coincidence degree of the output of virtual power plant resources and line load during the heavy - load period is to the maximum line overload value, the higher the reward, or it can also mean that the closer the average value of the output of virtual power plant resources during the heavy - load period is to the line average power, the higher the reward. The virtual power plant resource matching degree reward is expressed as follows: (14) Among them, represents the situation where the closer the coincidence degree of the output of virtual power plant resources and line load during the heavy - load period is to the maximum line overload value, the higher the reward, represents the situation where the closer the average value of the output of virtual power plant resources during the heavy - load period is to the line average power, the higher the reward; represents the maximum value of the line load power, represents the average value of the line load power.

[0046] Furthermore, after obtaining the response delay reward and the virtual power plant resource matching degree reward, a final reward function is established, which is expressed as follows: (15) Among them, respectively represent the weights corresponding to the response delay reward and the virtual power plant resource matching degree reward, represents the final reward.

[0047] As an example, in step S35, at a certain moment t when the simulation starts, the agent interacts with the distribution network environment in real - time to obtain the state space of the environment at this moment , and through calculation, the value function value of the action space corresponding to the state space is obtained .

[0048] (16) Among them, represents the weight matrix of the virtual power plant intelligent matching model, Q represents the value obtained by making an action selection according to the action (resource selection) - value function; represents the corresponding action space, represents the action space at the next moment, represents the state at the next moment; after executing the action , the distribution network environment changes, and the agent interacts with the distribution network environment to obtain the reward function value at the current moment, and judges the revenue value of the executed action based on the reward function value at the current moment, and takes Actions taken in the state The optimal return obtained is defined as the optimal state value function , which is expressed as follows: (17) Furthermore, the weight matrix during the training process is updated through the loss function and determined by the gradient descent method, where the loss function is expressed as follows: (18) where represents the loss function during the training process, and the calculation of the weight update during the training process is expressed as follows: (19) where represents the learning rate during the training process, which controls the update step size. According to the above training process, the above training steps are repeated, and the Q value is continuously updated until the training converges to obtain the optimal action, that is, the optimal virtual power plant resource selection combination.

[0049] In step S4, refer to Figure 1 step S4 in

[0050] to determine whether the value function converges. If it does, go to step S6; if not, go to step S5. (20) where represents Q the rate of change of the T value, represents the number of samples, k represents the weight matrix of the th training cycle, k-1 represents the weight matrix of the th training cycle. When

[0051] Figure 1 In step S5, refer to Figure 1 step S5 in to send the virtual power plant resource selection combination obtained during the training process to the optimal power flow calculation model. The optimal power flow calculation model aims to minimize the price of the virtual power plant resources used and returns the line load power obtained from the optimal power flow calculation of the distribution network after using the virtual power plant resource selection combination to the virtual power plant intelligent matching model to obtain the state space after the distribution network environment is updated

[0052] As an example, the optimal power flow calculation model aims to minimize the price of the virtual power plant resources used. According to the virtual power plant resource selection combination generated during the training process, it obtains the changes that occur after applying the virtual power plant resource selection combination generated during the training process to the distribution network environment, obtains the optimal power flow calculation results of the distribution network after using the virtual power plant resource selection combination, and obtains the set of optimized virtual power plant resource selection combinations. It sends the line load power obtained from the optimal power flow calculation results back to the virtual power plant intelligent matching model to optimize the matching results of the virtual power plant intelligent matching model.

[0053] As an example, the optimal power flow calculation model is used to calculate the power flow information of the distribution network, and the objective function is to minimize the price of the virtual power plant resources used. To ensure the safe and stable operation of the distribution network, considering that the branch power flow of the distribution network needs to satisfy the balance constraint, a constraint const 1.1 is set to ensure that the node voltages of the distribution network operate within a reasonable range during each period. const 1.2. The objective function and the constraint conditions const 1.1, the constraint conditions const 1.2 are expressed as follows: (21) Among them, a , b are the node numbers of the low-voltage distribution network; s is the node number of the low-voltage distribution network connected to the medium-voltage distribution network. The subscript contains s indicating that the parameter belongs to the low-voltage distribution network parameter of the medium-voltage distribution network node s ; , are the voltage values of nodes a , b at t periods; is t the active power flowing through the head end of branch ab at , are the impedances of branch ab ; is t the reactive power flowing through the head end of branch ab at indicating t the current flowing through branch ab at , are respectively t the active power and reactive power injected into node b at , respectively t instantaneous branch bk the active power and reactive power at the head end; k: b → k represented as b the set of all child nodes with the node as the parent node; and are respectively the upper limit value and the lower limit value of the node voltage of the low - voltage distribution network; represents the set of selected combination of virtual power plant resources used, represents the price of the i - th virtual power plant resource.

[0054] Furthermore, by performing optimal power flow calculation on the above - mentioned optimal power flow calculation model, the obtained optimal power flow calculation results include: line current, node voltage, and line load power.

[0055] Among them, t time - period branch ab the current flowing through is used to represent the line current; the node a and b at t the voltage value during the time - period and is used to represent the node voltage; t time - period branch ab the active power flowing through the head end is used to represent the line load power, and the line load power is updated.

[0056] In step S6, refer to Figure 1 step S6 in , and obtain the optimal intelligent matching combination of virtual power plant resources.

[0057] As an example, the optimal intelligent matching combination of virtual power plant resources obtained when training converges is represented as follows: (22) Among them, represents the optimal virtual power plant resource selection action, that is, the optimal intelligent matching combination of virtual power plant resources. This optimal intelligent matching combination of virtual power plant resources is the virtual power plant resource usage plan for virtual power plant optimal scheduling of the distribution network to be obtained in this application.

[0058] Please refer to Figure 2 and Figure 2 is the structure diagram of the deep Q - network method adopted in this application. Based on the deep Q - network method for virtual power plant resource intelligent matching optimization, it includes: Obtain the virtual power plant resource state at the current moment from the virtual power plant resource environment, and extract and transform the data required for the observation state; Input the data required for the extracted observation state into the deep Q-network neural network. The deep Q-network neural network calculates and outputs the value function Q, and selects actions through the greedy strategy, that is, select virtual power plant resources. Combine the selected virtual power plant resources into a virtual power plant resource selection combination, and send the virtual power plant resource selection combination to the distribution network environment according to the real-time state of the virtual power plant resource selection combination. The distribution network performs optimal power flow calculation to obtain the environmental state change after the action is executed. Select the virtual power plant resources, that is, select virtual power plant resources, combine the selected virtual power plant resources into a virtual power plant resource selection combination, and send the virtual power plant resource selection combination to the distribution network environment according to the real-time state of the virtual power plant resource selection combination. The distribution network performs optimal power flow calculation to obtain the environmental state change after the action is executed. After the action, the environmental state changes; Obtain the virtual power plant resource state variables from the distribution network environment , and form Deposit them into the experience pool. Among them, Y represents the state space, A represents the action space, R represents the reward, represents the updated state space. When the experience pool accumulates to the specified preset scale, randomly extract a certain amount of training data from the experience pool for training the deep Q-network neural network, and update the weight value of the deep Q-network neural network to continuously make the output value of the deep Q-network neural network close to the target value. Repeat this optimal scheduling method until the virtual power plant intelligent matching model converges to obtain the optimal control strategy, that is, the optimal virtual power plant resource intelligent matching combination.

[0059] This application also provides an artificial intelligence-based virtual power plant intelligent matching and optimal scheduling system for executing the above artificial intelligence-based virtual power plant intelligent matching and optimal scheduling method. Please refer to Figure 3 , Figure 3 is the module diagram of the artificial intelligence-based virtual power plant intelligent matching and optimal scheduling system provided by this application. The artificial intelligence-based virtual power plant intelligent matching and optimal scheduling system provided by this application includes: The virtual power plant resource calculation module 31 is used to construct an aggregated analysis model of the virtual power plant considering new energy with the minimum expected value of the power shortage of all new energy in the distribution network as the objective function, and calculate all virtual power plant resources in the distribution network through the particle swarm optimization algorithm; The virtual power plant resource measurement index establishment module 32 is used to establish virtual power plant resource measurement indexes according to all corresponding virtual power plant resources in the distribution network; The virtual power plant resource intelligent matching module 33 is used to establish a virtual power plant intelligent matching model based on the deep Q-network method according to the established virtual power plant resource measurement indexes, and train the virtual power plant intelligent matching model to obtain the value function; judge whether the value function converges. If so, obtain the optimal virtual power plant resource intelligent matching combination; if not, send the virtual power plant resource selection combination obtained during the training process to the optimal power flow calculation model, obtain the updated state space of the environment and train again until the value function converges; The virtual power plant combination optimization module 34 is used to construct an optimal power flow calculation model with the goal of minimizing the price of the virtual power plant resources used, and return the line load obtained from the optimal power flow calculation of the distribution network after the virtual power plant resource selection combination to the virtual power plant intelligent matching model.

[0060] As an example, the virtual power plant resource calculation module 31, the virtual power plant resource measurement index establishment module 32, the virtual power plant resource intelligent matching module 33, and the virtual power plant combination optimization module 34 are connected in sequence.

[0061] The present application also provides a computer program product, including a computer program, which when executed by a processor implements the steps of the artificial intelligence-based virtual power plant intelligent matching and optimal scheduling method described in any one of the above.

[0062] The artificial intelligence-based virtual power plant intelligent matching and optimal scheduling method, system, and computer program product provided by the present application construct a virtual power plant intelligent matching model through the deep Q-network method to solve the deficiencies of the prior art in dealing with new energy volatility, load uncertainty, and resource matching optimization, and have significant advantages in dynamic adaptability, accuracy, and economy. Through the self-learning ability of reinforcement learning, the system can autonomously optimize the scheduling strategy, dynamically adapt to the fluctuations of new energy output and load changes, and overcome the defects of relying on manual experience and being difficult to cope with uncertainties. At the same time, the present application not only considers the maximum output capacity of virtual power plant resources, but also comprehensively introduces measurement indexes such as line load matching degree and response delay, making the resource scheduling more accurate, thereby effectively reducing the line overload rate and improving the safety and stability of the power grid. In addition, combined with the optimal power flow calculation model, with the goal of minimizing the use cost of the virtual power plant, while ensuring the safe operation of the power grid, the economy is optimized, providing a more efficient and flexible solution for new energy consumption and intelligent power grid scheduling.

[0063] The present application can not only improve the safety and economy of the distribution network operation, but also improve the new energy consumption capacity, enhance the flexibility of the power grid regulation, and provide an intelligent solution for the efficient utilization of large-scale distributed energy.

[0064] Although the present application has been disclosed as above with embodiments, it is not intended to limit the present application. Any person with ordinary knowledge in the technical field to which the present application pertains may make some modifications and refinements without departing from the spirit and scope of the present application. Therefore, the protection scope of the present application shall be subject to that defined by the appended patent application scope.

Claims

1. An intelligent matching and optimal scheduling method for a virtual power plant based on artificial intelligence, characterized in that, including, Step S1: Taking the minimum expected value of the power shortage of all new energy in the distribution network as the objective function, a virtual power plant aggregation analysis model considering new energy is constructed, and all virtual power plant resources in the distribution network are calculated through the particle swarm optimization algorithm; Step S2: Based on all virtual power plant resources in the distribution network, a virtual power plant resource measurement index is established; Step S3: Based on the established virtual power plant resource measurement index, a virtual power plant intelligent matching model is established based on the deep Q-network method, and the value function is obtained by training the virtual power plant intelligent matching model; Step S4: Determine whether the value function converges. If it does, go to Step S6; if not, go to Step S5; Step S5: Send the virtual power plant resource selection combination obtained during the training process to the optimal power flow calculation model. The optimal power flow calculation model aims to minimize the price of the virtual power plant resources used, and returns the line load power obtained by calculating the optimal power flow of the distribution network after using the virtual power plant resource selection combination to the virtual power plant intelligent matching model, obtains the state space after the distribution network environment is updated, and trains again, returning to Step S4; Step S6: Obtain the optimal virtual power plant resource intelligent matching combination.

2. The intelligent matching and optimal scheduling method of the virtual power plant based on artificial intelligence according to claim 1, characterized in that, The objective function of the virtual power plant aggregation analysis model is expressed as follows: Among them, represents the objective function that minimizes the expected value of the power shortage of all new energy sources in the distribution network, E represents the expected value operator, D represents the load demand of the distribution network, represents the total output of all new energy sources in the distribution network; The constraint conditions of the virtual power plant aggregation analysis model include the output constraints of photovoltaic resources and wind power resources, and are expressed as follows: Among them, represents the i-th photovoltaic resource, represents the j-th wind power resource, N, M respectively represent the quantities of photovoltaic resources and wind power resources; and respectively represent the upper and lower limits of the output of photovoltaic resources, and respectively represent the upper and lower limits of the output of wind power resources.

3. The intelligent matching and optimal scheduling method of the virtual power plant based on artificial intelligence according to claim 1, characterized in that The virtual power plant resource measurement index includes: line load and the response latency of virtual power plant resources, the coincidence degree of the output of virtual power plant resources and line load during heavy load periods, and the average value of the output of virtual power plant resources during heavy load periods; The line load and the response latency of virtual power plant resources are expressed as follows: Among them, S1 is the response latency of line load and virtual power plant resources, indicating the overall correlation coefficient between the i th line load and the j th virtual power plant resource, indicating the i th line load, indicating the jth virtual power plant resource; The coincidence degree of the output of virtual power plant resources and line load during heavy load periods is expressed as follows: Among them, S 2 is the second virtual power plant resource measurement index, represents the heavy-load period of the i th line, represents the value corresponding to the load of the i-th line during the heavy-load period, represents the value corresponding to the j-th virtual power plant resource during the heavy-load period, represents the weight used to calculate the coincidence degree of output; The average value of the output of virtual power plant resources during heavy load periods is expressed as follows: Among them, S 3 is the third virtual power plant resource measurement indicator, indicating the heavy-load period of the i-th line.

4. The intelligent matching and optimal scheduling method of a virtual power plant based on artificial intelligence according to claim 1, characterized in that, Step S3 includes: Obtain the current virtual power plant resource state from the virtual power plant resource environment and extract the required data; Establish the state space of the deep Q-network method according to the virtual power plant resource measurement index; Establish the action space of the deep Q-network method; Design the reward function of the deep Q-network method according to the virtual power plant resource measurement index; Train the established deep Q-network model to obtain the corresponding action value function.

5. The intelligent matching and optimal scheduling method of a virtual power plant based on artificial intelligence according to claim 4, characterized in that, The required data extracted includes: line load power, line rated power, and line power during heavy load periods; The corresponding action value function is expressed as follows: Among them, represents the weight matrix of the virtual power plant intelligent matching model, represents the value obtained by making action selections based on the action-value function; represents the state-action space at time t, A t represents Y t the corresponding action space, represents the action space at the next moment, represents the state at the next moment.

6. The intelligent matching and optimal scheduling method of the virtual power plant based on artificial intelligence according to claim 5, characterized in that, The judgment formula for the convergence of the value function obtained by training is expressed as follows: Among them, represents Q the value change rate, T represents the number of samples in the experience pool, represents the k weight matrix of the th training cycle, k-1 represents the weight matrix of the th training cycle; when training converges, otherwise, it does not converge; represents the convergence judgment parameter.

7. The intelligent matching and optimal scheduling method of a virtual power plant based on artificial intelligence according to claim 1, characterized in that The objective function of the optimal power flow calculation model, the first constraint condition of the optimal power flow calculation model, and the second constraint condition of the optimal power flow calculation model are expressed as follows: Among them, represents the objective function of the optimal power flow calculation model, const 1.1 represents the first constraint condition of the optimal power flow calculation model, const 1.2 represents the second constraint condition of the optimal power flow calculation model; a 、 b is the node number of the low-voltage distribution network; s is the node number of the low-voltage distribution network connected to the medium-voltage distribution network, and the subscript contains s indicating that the parameter belongs to the low-voltage distribution network parameter of the medium-voltage distribution network node s ; 、 are the voltage values of nodes a 、 b at time t ; is t the active power flowing through the head end of branch ab at time 、 are the impedance of branch ab ; is t the reactive power flowing through the head end of branch ab at time represents t the current flowing through branch ab at time 、 are respectively t the active power and reactive power injected into node b at time 、 are respectively t the active power and reactive power at the head end of branch bk at time k:b→k represents the set of all child nodes with b node as the parent node; 、 are respectively the upper limit value and lower limit value of the low-voltage distribution network node voltage; represents the set of the selected combination of virtual power plant resources used, represents the price of the i-th virtual power plant resource.

8. The intelligent matching and optimal scheduling method of the virtual power plant based on artificial intelligence according to claim 1, characterized in that, The optimal virtual power plant resource intelligent matching combination is expressed as follows: Among them, represents the optimal virtual power plant resource selection action, represents the optimal state value function.

9. An intelligent matching and optimal scheduling system for a virtual power plant based on artificial intelligence, which is used to execute the intelligent matching and optimal scheduling method for a virtual power plant based on artificial intelligence as described in any one of claims 1 to 8, characterized in that including: a virtual power plant resource calculation module, a virtual power plant resource measurement index establishment module, a virtual power plant resource intelligent matching module, and a virtual power plant combination optimization module; The virtual power plant resource calculation module, the virtual power plant resource measurement index establishment module, the virtual power plant resource intelligent matching module, and the virtual power plant combination optimization module are connected in sequence.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the artificial intelligence-based intelligent matching and optimal scheduling method for a virtual power plant as described in any one of claims 1 to 8.

Images