Virtual power plant multi-level collaborative optimization control method based on digital twinning

Abstract

The invention relates to the technical field of intelligent power grids, in particular to a virtual power plant multi-level collaborative optimization control method based on digital twinning. According to the invention, data of the power generation end, the power storage end, the power transmission end and the load end are collected and processed in real time, a digital twin model is established, and accurate modeling and real-time monitoring can be carried out on each device of the virtual power plant. Through real-time optimization scheduling and cooperative control, energy loss can be minimized, supply and demand can be balanced, the energy use efficiency is improved to the maximum extent, and the cost is reduced; according to the method, the genetic algorithm and the particle swarm optimization algorithm are combined, scheduling optimization is carried out based on a multi-level collaborative optimization model, a virtual power plant can rapidly respond according to changes of real-time environment and load requirements, fossil energy reduction and clean energy increase are preferentially considered in the power generation equipment scheduling process, and greenhouse gas emission reduction is facilitated.

Description

Technical Field

[0001] The present invention relates to the technical field of smart grids, and specifically to a multi-level collaborative optimization control method for virtual power plants based on digital twins. Background Art

[0002] Smart grid technology is a power network system that uses information technology, communication technology, control technology, and automation technology to achieve dynamic monitoring, optimized dispatching, and intelligent management of the power system. The smart grid is not only an upgraded version of the traditional power grid but also enables real-time monitoring and control of the energy flow, optimizes the allocation of power resources, and improves the reliability, flexibility, sustainability, and economy of the power grid. Although certain progress has been made in current power system and virtual power plant management technologies, there are still multiple problems and deficiencies, especially in terms of optimized dispatching and resource allocation, specifically including the following points: First, the energy utilization efficiency is low, and renewable energy cannot be fully utilized. Traditional power system dispatching mostly relies on static models and is difficult to dynamically adjust the output of various energy sources, especially in the utilization of renewable energy such as wind energy and photovoltaic energy. Due to weather changes and prediction errors, the volatility of clean energy is very large, and the traditional system cannot accurately control the real-time balance between the power generation side and the load side, resulting in the underutilization of renewable energy.

[0003] Second, traditional power grid dispatching systems mostly rely on static dispatching algorithms and often do not consider seasonal fluctuations in power demand, emergencies, or equipment failures. When encountering load fluctuations or equipment failures, the power grid may experience problems such as supply-demand imbalance, overload, or power outage, leading to a decline in the security and stability of the power grid.

[0004] Finally, when traditional power grid dispatching performs optimization, it only considers the optimization of the supply side and the load side and does not consider the significant energy losses that occur during long-distance power transmission. These losses are often difficult to accurately predict and optimize, resulting in a low overall efficiency of the power system and an increase in operating costs.

[0005] In view of the above problems, it is necessary to propose a multi-level collaborative optimization control method for virtual power plants based on digital twins. Summary of the Invention

[0006] The purpose of the present invention is to solve the problems existing in the background art and propose a multi-level collaborative optimization control method for virtual power plants based on digital twins.

[0007] The purpose of the present invention can be achieved through the following technical solutions: A multi-level collaborative optimization control method for virtual power plants based on digital twins includes the following steps: Step 1: Data collection and digital twin model construction; The operation data of each device from the power generation end, transmission end, power storage end to the load end of the virtual power plant is collected in real time through a sensor group and an information acquisition device; the environmental data is collected in real time from an environmental data sensor.

[0008] The specific operation data is as follows: Operation data of the power generation end , including the serial number ID of the power generation equipment, output power, production volume of fossil energy input, real-time power generation volume and historical power generation volume; the power generation equipment includes: traditional fossil energy power generation equipment, photovoltaic power generation equipment and wind power generation equipment. For the photovoltaic power generation equipment and wind power generation equipment, the corresponding production volume of fossil energy input is limited to 0.

[0009] Operation data of the transmission end , including the serial number ID of the transmission equipment, the serial number ID of the upstream equipment connected to the transmission equipment and the serial number ID of the downstream equipment, voltage, current, power loss and load rate; among them, the upstream equipment and the downstream equipment include any combination of two of the power generation equipment, energy storage equipment and load end equipment.

[0010] Operation data of the power storage end , including the serial number ID of the energy storage equipment, remaining battery power, input power, output power and charging efficiency; Operation data of the load end equipment , including the serial number ID of the load end equipment, load demand, weather forecast data, grid frequency and voltage.

[0011] The specific environmental data is as follows: Environmental parameters , including the real-time wind speed, temperature and light intensity in the areas where the power generation equipment, energy storage equipment and load end equipment are located.

[0012] As a preferred mode of the present invention, a real-time simulation model of the virtual power plant is established through digital twin technology, and the collected operation data and environmental data are input into the real-time simulation model. The model includes the physical entities, operation states, operation data, environmental data, energy flow and cooperation relationships between devices of the virtual power plant.

[0013] Step two: Establish a multi-level collaborative optimization model; Perform multi-level classification in the real-time simulation model of the virtual power plant, including: Level one: The real layer, including the power generation equipment, transmission equipment, energy storage equipment and load end equipment in each power generation end, transmission end, power storage end and load end, as well as the operation data and environmental data of each device collected, and the input data of level one is the real data set ; Level 2: Virtual layer, including the power generation prediction model at the power generation end, the topological structure model at the power transmission end, the scheduling and allocation model at the power storage end, and the load prediction model at the load end; Level 3: Scheduling optimization layer, including several scheduling optimization signals generated based on the superposition and comparison of the real layer and the virtual layer; In the above-mentioned Level 2: Virtual layer, the power generation prediction model at the power generation end, the topological structure and transmission loss model at the power transmission end, the scheduling and allocation model at the power storage end, and the load prediction model at the load end are specifically as follows: Power generation prediction model: It includes the fossil energy power generation prediction sub-model, the photovoltaic power generation prediction sub-model, and the wind power generation prediction sub-model. The mathematical model formula is: The formula is: ; where is the predicted total output power value of traditional fossil energy power generation equipment at time t; where is the predicted total output power value of photovoltaic power generation equipment at future time t; where is the predicted total output power value of wind power generation equipment at future time t; where is the predicted total power generation value at future time t; where and are the support vector regression functions trained for traditional fossil energy power generation equipment, photovoltaic power generation equipment, and wind power generation equipment respectively, used to predict the traditional fossil energy, photovoltaic, and wind power generation at future time t; where is the planned production volume of fossil energy at time t, is the total output power of traditional fossil energy power generation equipment at the current moment, is the predicted average temperature of the power generation equipment at future time t; where is the predicted light intensity value at future time t, is the total output power of photovoltaic power generation equipment at the current moment; where is the predicted wind speed value at future time t, is the total output power of wind power generation equipment at the current moment; where and are the prediction errors of the fossil energy power generation prediction sub-model, the photovoltaic power generation prediction sub-model, and the wind power generation prediction sub-model respectively, following a Gaussian distribution; where is the predicted total output power value of all equipment at the power generation end at future time t.

[0014] The topological structure and transmission loss model at the power transmission end are specifically as follows: Where i1 and i2 are respectively the ID numbers of the upstream device and the downstream device connected to each power transmission device, and I is the set composed of the ID numbers of all power generation devices, energy storage devices, and load devices. Among them, C(i1, i2) is the power transmission dispatching symbol. A C(i1, i2) value of 1 represents power transmission work between the upstream device i1 and the downstream device i2; a C(i1, i2) value of 0 represents no power transmission work between the upstream device i1 and the downstream device i2. Among them represents the power transmission power between the upstream device i1 and the downstream device i2. Among them, V(i1, i2) is the transmission line voltage between the upstream device i1 and the downstream device i2.

[0015] The scheduling and allocation model of the electricity storage end is specifically as follows: ; Among them is the predicted electricity storage capacity of all energy storage devices at time t. Among them is the actual electricity storage capacity of the energy storage device at the previous historical time t; Among them is the total input power and total output power of all energy storage devices collected at time x; Among them are the charge and discharge efficiencies respectively; Among them is the predicted management cost generated by the charge and discharge behavior, are the management costs generated by the charge and discharge scheduling per unit power respectively; Among them and are the maximum threshold of the total input power, the maximum threshold of the total output power, and the maximum threshold of the electricity storage capacity of all energy storage devices respectively.

[0016] The load prediction model of the load end: Through the trained LSTM long short-term memory network, including the forget gate, input gate, candidate memory unit, update memory unit, and output gate, the total input of its input gate is historical load data, the real-time wind speed, temperature, and light intensity in the area where the load device is located, and special event dates including holidays and weekdays, and the output is the predicted value of the total load power at the future time t .

[0017] Among them, the core operation formula of the LSTM long short-term memory network is: Forget gate: ; Among them is the output value of the forget gate, is the Sigmoid activation function; Among them is the hidden state at the previous moment; Among them is the input at the current moment, that is, the data vector composed of historical load data, the real-time wind speed, temperature, and light intensity in the area where the load device is located, and special event dates including holidays and weekdays; Among them They are the weight term and bias term of the forget gate respectively.

[0018] Input gate: ; where is the output value of the input gate, where They are the weight term and bias term of the input gate respectively.

[0019] Candidate memory gate: ; where is the output value of the candidate memory gate; tanh is the hyperbolic tangent activation function, They are the weight term and bias term of the candidate memory gate respectively.

[0020] Update memory cell: ; where is the output value of the updated memory cell; where is the output value of the updated memory cell at the previous moment; Output gate: ; where is the output value of the output gate, is the hidden state at the current moment, and after a fully connected operation on all the total output of the LSTM long short-term memory network is obtained; They are the weight term and bias term of the output gate respectively.

[0021] As a preferred embodiment of the present invention, the multi-objective optimization function of the second layer is set, including: Minimize the energy loss objective optimization function: , that is, minimize the total power loss at the power transmission end; Balance the supply and demand objective optimization function: , that is, the total output power of the power generation end and the energy storage end is equal to the total output power of the load end; Minimize the power generation cost objective optimization function: , that is, minimize the cost generated by the fossil energy production and the energy storage end scheduling management, where is the cost required for the planned production of unit fossil energy; Minimize the carbon emission objective optimization function: , that is, minimize the planned production volume of fossil energy at the power generation end.

[0022] Step three: Data processing and collaborative optimization; Weight coefficients λ1, λ2, λ3, and λ4 are respectively assigned to the optimization function for minimizing energy loss, the optimization function for balancing power supply and demand, the optimization function for minimizing power generation cost, and the optimization function for minimizing carbon emissions, which respectively represent the importance of each multi-objective optimization function. The sum of λ1, λ2, λ3, and λ4 is 1, and their specific values are dynamically adjusted according to dispatching requirements. The optimization functions for minimizing energy loss, balancing power supply and demand, minimizing power generation cost, and minimizing carbon emissions are weighted and summed according to their respective assigned weight coefficients to obtain the final optimization objective function F.

[0023] The specific form of the optimization objective function F is as follows: .

[0024] Furthermore, whenever an optimization operation instruction is received from the administrator, a joint optimization model based on the GA genetic algorithm and the PSO particle swarm optimization is built for multi-objective optimization operations, and the final optimization objective function F is solved. Through selection, crossover, and mutation operations, a virtual power plant optimization control scheme that satisfies the condition of minimizing the final optimization objective function F is obtained. The specific process is as follows: Initialize the population particles of GA and PSO. Initialize the GA population particles to generate a set of GA population particles representing initial candidate solutions, where each candidate solution represents a control scheme of the virtual power plant; each particle represents a control scheme of the virtual power plant. For the GA population particles, perform selection, crossover, and mutation operations, and evaluate the fitness of each GA population particle according to the specific value of the objective function F. Among the GA population particles, the smaller the objective function F of a specific particle, the higher the fitness of the particle, and the more likely it is to be selected into the next generation. Calculate the probability P of each particle being selected into the next generation by operating on the objective function F of each particle in the GA population particles. And perform crossover operations on all the particles in the GA population particles. Extract a preset number of particles according to the probability P of each particle being selected into the next generation, and randomly exchange the specific values of each dimension in the multi-dimensional vectors corresponding to the selected particles through crossover operations to simulate partial gene exchange and obtain the new solution after crossover operations. Convert the new solution obtained after the crossover operation into PSO population particles and initialize them to generate a set of initial PSO population particles, where each particle represents a control scheme of the virtual power plant. Input all the PSO population particles obtained by initializing the new solution after the crossover operation into the PSO velocity update and position update formulas for PSO population optimization.

[0025] Among them, the GA population particles and the PSO population particles are both multi-dimensional vectors, and each dimension corresponds to the real-time output power of each power generation device at each moment t, the planned production volume of fossil energy at each moment t, the real-time output power of the power generation device at each moment t, the power transmission scheduling symbol C(i1, i2) of all upstream devices i1 and downstream devices i2 at each moment t, and the power transmission power between any upstream device i1 and downstream device i2 at each moment t the total input power and total output power of the energy storage device at each moment t, and the power consumption limit of the load end at each moment t.

[0026] The PSO speed update and position update formulas are as follows: ; where is the update formula for the particle velocity, where is the velocity of particle k in the (q + 1)-th operation after update; where k is the particle number index and q is the operation iteration number index; where w is the preset inertia weight, which controls the influence degree of the particle's historical velocity on the updated velocity in the next operation, where is the velocity of particle k in the q-th operation; where c1 and c2 are the preset learning factors, which respectively control the dependence degrees of the particle on the personal best position and the global best position; where r1 and r2 are random numbers between [0, 1]; where is the position of particle k in the q-th operation, that is, the multi-dimensional vector representing the virtual power plant control scheme; where is the personal optimal solution of particle k, that is, the particle position where the objective function F is minimized for particle k in the iterative operation; where is the global optimal solution, that is, the particle position where the objective function F is minimized for all particles in the iterative operation; where is the particle position update formula.

[0027] Let the stopping conditions for the iterative operation of PSO population optimization be: Condition 1: The iteration number q is greater than the preset upper limit qmax of the optimization operation times; Condition 2: The updated velocities of all particles k converge; When it is recognized that either Condition 1 or Condition 2 is satisfied, it is determined that the optimization is successful, the iterative operation of the PSO speed update and position update formulas is stopped, and the global optimal solution is output as the final optimization result; Obtain the multi-dimensional vector corresponding to the global optimal solution, and match the parameter explanations corresponding to each dimension to obtain the virtual power plant optimization control scheme.

[0028] As a preferred embodiment of the present invention, whenever an optimization operation instruction sent by the administrator is received, a joint optimization model based on the GA genetic algorithm and the PSO particle swarm optimization is established for multi-objective optimization operation, and the final optimization objective function is solved. Through selection, crossover, and mutation operations, a virtual power plant optimization control scheme that satisfies the condition that the final optimization objective function F is minimized is obtained.

[0029] The virtual power plant optimization control scheme described above includes: The scheduling strategy of power generation equipment, that is, the real-time output power of power generation equipment at each moment t and the planned production volume of fossil energy at each moment t; The scheduling strategy of the transmission network, that is, the transmission scheduling symbol at each moment t; The scheduling strategy of energy storage equipment, that is, the total input power and total output power of energy storage equipment at each moment t.

[0030] The scheduling strategy of the load side, that is, the power limit of the load side power consumption at each moment t.

[0031] Step Four: Cooperative optimization signal matching; As a preferred embodiment of the present invention, based on the virtual power plant optimization control scheme, the corresponding control signals are matched and saved to Level Three: Scheduling Optimization Layer. The control signals include: Power generation end power adjustment signal: The signal for increasing / decreasing the production volume of fossil energy at each moment t, and the target quality of the increase / decrease; Energy storage end control signal: The charge / discharge instruction of the energy storage end at each moment t, and the target power of the charge / discharge of the energy storage end; Transmission end control signal: The instruction to turn on / off the transmission end equipment at each moment t; Load side load adjustment signal: The power limit instruction of the load side of the power grid scheduling at each moment t, and the target power of the restricted load side.

[0032] The control signals at each moment t saved in Level Three: Scheduling Optimization Layer are sent to the management platform for manual review. After obtaining the confirmation of the manual review, they are output to Level One: Reality Layer for control signal execution.

[0033] Step Five: Optimization control strategy execution and supervision; During the execution of the control signal, the execution effect is jointly supervised through a PID supervisor and an objective function. The specific process is as follows: Obtain the target quality of increasing / decreasing the production volume of fossil energy included in the power generation end power adjustment signal; Every preset time interval t, calculate the difference between the target quality of increasing / decreasing and the actual increase / decrease volume to obtain the first execution deviation.

[0034] Obtain the target power of charging / discharging included in the control signal of the electricity storage end; every preset time interval t, calculate the difference between the target power of charging / discharging and the actual charging / discharging power to obtain the second execution deviation.

[0035] The target power of the load end included in the load regulation signal of the load end; every preset time interval t, calculate the difference between the target power of the load end and the actual load end power to obtain the third execution deviation.

[0036] Take the average value of the first, second, and third execution deviations as the total execution deviation e(t) and input it into the PID supervisor: Calculate the PID deviation eigenvalue u(t); where They are respectively the preset weight coefficients of proportional, integral, and differential; where t1 is the generation time of the control signal, that is, the time when the control instruction is output after operation after receiving the optimization operation instruction sent by the administrator last time.

[0037] When it is recognized that the PID deviation eigenvalue u(t) is greater than the first preset threshold, or it is recognized that the increasing speed of the final optimization objective function F with time t is greater than the second preset threshold, it is determined that there is a huge deviation between the control signal and the actual control effect, and execute the isolation between layer one: the reality layer and layer three: the scheduling optimization layer, and stop the generation of all new control signals and the execution of existing control signals.

[0038] Compared with the prior art, the beneficial effects of the present invention are: 1. By collecting and processing data of the power generation end, electricity storage end, power transmission end, and load end in real time, the present invention establishes a digital twin model, which can accurately model and monitor each device of the virtual power plant. Through real-time optimization scheduling and coordinated control, it can not only minimize energy loss, but also balance supply and demand, maximize energy use efficiency, and reduce costs. For example, by precisely adjusting the output of power generation equipment and the charging / discharging strategy of energy storage equipment, the overall performance and economy of the power system are optimized; 2. The present invention adopts a scheduling control scheme based on multi-objective optimization, in which the goal of minimizing carbon emissions is a key optimization goal, which can effectively reduce the use of fossil energy in the virtual power plant and reduce the emissions of pollutants such as carbon dioxide. During the scheduling process of power generation equipment, the system will give priority to using clean energy and reduce carbon emissions by controlling the increase or decrease of fossil energy power generation. This method not only helps to reduce greenhouse gas emissions, but also supports the virtual power plant to develop towards a green and sustainable energy structure; 3. The present invention combines a genetic algorithm and a particle swarm optimization algorithm, and performs scheduling optimization based on a multi-level collaborative optimization model, enabling the virtual power plant to respond quickly according to changes in the real-time environment and load demand. Through the intelligent decision-making system, the system can dynamically adjust the external environment and the internal equipment status. This flexible adaptive ability can effectively handle emergencies, ensure the stable operation of the virtual power plant, and make the best scheduling decisions in the face of complex power demands. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] For the convenience of those skilled in the art to understand, the present invention will be further described below with reference to the accompanying drawings: Figure 1 is the flowchart of the method of the present invention; Figure 2 is the topological structure diagram of the LSTM long short-term memory network proposed in the embodiment of the present invention; DETAILED DESCRIPTION OF THE EMBODIMENTS

[0040] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0041] Please refer to Figure 1 shown, the multi-level collaborative optimization control method of the virtual power plant based on digital twin includes the following steps: Step 1: Data acquisition and digital twin model construction; The operating data of each device in the virtual power plant from the power generation end, transmission end, energy storage end to the load end is collected in real time through a sensor group and information collection equipment; the environmental data is collected in real time from the environmental data sensor.

[0042] The specific operating data is as follows: The operating data of the power generation end , including the ID number of the power generation equipment, output power, fossil energy input, real-time power generation and historical power generation; the power generation equipment includes: traditional fossil energy power generation equipment, photovoltaic power generation equipment and wind power generation equipment. For the photovoltaic power generation equipment and wind power generation equipment, the corresponding fossil energy input is limited to 0.

[0043] The operating data of the transmission end , including the ID number of the transmission equipment, the ID number of the upstream equipment connected to the transmission equipment and the ID number of the downstream equipment, voltage, current, power loss and load rate; among them, the upstream equipment and the downstream equipment include any combination of two of the power generation equipment, energy storage equipment and load end equipment.

[0044] Operating data of the electricity storage end , including the serial number ID of the energy storage device, remaining battery power, input power, output power, and charging efficiency; Operating data of the load end device , including the serial number ID of the load end device, load demand, weather forecast data, grid frequency, and voltage.

[0045] The specific environmental data is as follows: Environmental parameters , including the real-time wind speed, temperature, and light intensity in the areas where the power generation equipment, energy storage equipment, and load end equipment are located.

[0046] Furthermore, a real-time simulation model of the virtual power plant is established through digital twin technology, and the collected operating data and environmental data are input into the real-time simulation model. The real-time simulation model of the virtual power plant includes the physical entities, operating states, operating data, environmental data, energy flow, and cooperation relationships between devices of the virtual power plant.

[0047] Step 2: Establish a multi-level collaborative optimization model; Perform multi-level classification in the real-time simulation model of the virtual power plant, including: Level 1: Reality layer, including the power generation equipment, power transmission equipment, energy storage equipment, and load end equipment in each power generation end, power transmission end, electricity storage end, and load end, as well as the operating data and environmental data of each device collected. The input data of Level 1 is the real data set ; Level 2: Virtual layer, including the power generation prediction model at the power generation end, the topological structure model at the power transmission end, the scheduling and allocation model at the electricity storage end, and the load prediction model at the load end; Level 3: Scheduling optimization layer, including several scheduling optimization signals generated based on the superposition and comparison of the reality layer and the virtual layer; In the said Level 2: Virtual layer, the power generation prediction model at the power generation end, the topological structure and power transmission loss model at the power transmission end, the scheduling and allocation model at the electricity storage end, and the load prediction model at the load end are specifically as follows: Power generation prediction model: including the fossil energy power generation prediction sub-model, the photovoltaic power generation prediction sub-model, and the wind power generation prediction sub-model. The mathematical model formula is: The formula is: ; where is the predicted total output power value of the traditional fossil energy power generation equipment at time t; where is the predicted total output power value of the photovoltaic power generation equipment at future time t; where is the predicted total output power value of the wind power generation equipment at future time t; where is the predicted total power generation at future time t; where and are support vector regression functions trained for traditional fossil energy power generation equipment, photovoltaic power generation equipment, and wind power generation equipment respectively, and are used to predict the traditional fossil energy, photovoltaic, and wind power generation at future time t; where is the planned production volume of fossil energy at time t, is the total output power of traditional fossil energy power generation equipment at the current time, is the predicted average temperature of the power generation equipment at future time t; where is the predicted value of the light intensity at future time t, is the total output power of the photovoltaic power generation equipment at the current time; where is the predicted value of the wind speed at future time t, is the total output power of the wind power generation equipment at the current time; where and are the prediction errors of the fossil energy power generation prediction sub-model, photovoltaic power generation prediction sub-model, and wind power generation prediction sub-model respectively, and follow a Gaussian distribution; where is the predicted value of the total output power of all equipment at the power generation end at future time t.

[0048] The topological structure and transmission loss model at the power transmission end are specifically as follows: where i1 and i2 are the ID numbers of the upstream equipment and downstream equipment connected by each transmission equipment respectively, and I is the set composed of the ID numbers of all power generation equipment, energy storage equipment, and load end equipment. Among them, C(i1, i2) is the transmission scheduling symbol, and the value of C(i1, i2) being 1 represents that power transmission work is carried out between the upstream equipment i1 and the downstream equipment i2; the value of C(i1, i2) being 0 represents that power transmission work is not carried out between the upstream equipment i1 and the downstream equipment i2; where represents the transmission power between the upstream equipment i1 and the downstream equipment i2, where V(i1, i2) is the transmission line voltage between the upstream equipment i1 and the downstream equipment i2.

[0049] The scheduling and allocation model at the energy storage end is specifically as follows: ; where is the predicted stored electricity of all energy storage equipment at time t, where is the actual stored electricity of the energy storage equipment at the previous historical time t; where is the total input power and total output power of all energy storage equipment collected at time x; where are the charge and discharge efficiencies respectively; where is the predicted management cost generated by the charge and discharge behavior, The management costs generated by the charge and discharge scheduling of unit power respectively; among which and are the maximum threshold of the total input power, the maximum threshold of the total output power, and the maximum threshold of the stored electricity of all energy storage devices respectively.

[0050] Load-side load prediction model: Through the trained LSTM long short-term memory network, including the forget gate, input gate, candidate memory unit, update memory unit, and output gate, the total input of the input gate is historical load data, the real-time wind speed, temperature, and light intensity in the area where the load-side device is located, and special event dates including holidays and weekdays, and the output is the predicted value of the total load-side power at the future time t .

[0051] Please refer to Figure 2 As shown, the topological structure diagram of the LSTM long short-term memory network, where the core operation formula of the LSTM long short-term memory network is: Forget gate: ; among which is the output value of the forget gate, is the Sigmoid activation function; among which is the hidden state at the previous moment; among which is the input at the current moment, that is, the data vector composed of historical load data, the real-time wind speed, temperature, and light intensity in the area where the load-side device is located, and special event dates including holidays and weekdays; among which are the weight term and bias term of the forget gate respectively.

[0052] Input gate: ; among which is the output value of the input gate, among which are the weight term and bias term of the input gate respectively.

[0053] Candidate memory gate: ; among which is the output value of the candidate memory gate; tanh is the hyperbolic tangent activation function, are the weight term and bias term of the candidate memory gate respectively.

[0054] Update memory unit: ; among which is the output value of the update memory unit; among which is the output value of the update memory unit at the previous moment; Output gate: ; among which is the output value of the output gate, is the hidden state at the current moment, after all The total output of the LSTM long short-term memory network is obtained from the full link ; where are the weight term and bias term of the output gate respectively.

[0055] Furthermore, set the multi-objective optimization function of the second layer, including:[[]] Minimize the energy loss objective optimization function: , that is, minimize the total power loss at the transmission end; Balance the supply and demand objective optimization function: , that is, the total output power of the power generation end and the energy storage end is equal to the total output power of the load end; Minimize the power generation cost objective optimization function: , that is, minimize the cost generated by the commissioning of fossil energy and the scheduling management of the energy storage end, where is the cost required for the planned commissioning of unit fossil energy; Minimize the carbon emission objective optimization function: , that is, minimize the planned production volume of fossil energy at the power generation end.

[0056] Step 3: Data processing and collaborative optimization; Weight coefficients λ1, λ2, λ3, and λ4 are assigned to the minimize energy loss objective optimization function f1, balance supply and demand objective optimization function f2, minimize power generation cost objective optimization function f3, and minimize carbon emission objective optimization function f4 respectively, which represent the importance of each multi-objective optimization function. The sum of λ1, λ2, λ3, and λ4 is 1, and their specific values are dynamically adjusted according to scheduling requirements. Weighted summation is performed on the minimize energy loss objective optimization function f1, balance supply and demand objective optimization function f2, minimize power generation cost objective optimization function f3, and minimize carbon emission objective optimization function f4 according to their respective assigned weight coefficients to obtain the final optimization objective function F.

[0057] The specific form of the optimization objective function F is: .

[0058] Furthermore, whenever an optimization operation instruction sent by the administrator is received, a joint optimization model based on the GA genetic algorithm and PSO particle swarm optimization is built for multi-objective optimization operation, and the final optimization objective function F is solved. Through selection, crossover, and mutation operations, a virtual power plant optimization control scheme that meets the condition: the final optimization objective function F is minimized is obtained. The specific process is: Initialize the population particles of GA and PSO. Initialize the GA population particles to generate a set of GA population particles representing the initial candidate solutions, where each candidate solution represents a control scheme for the virtual power plant; each particle represents a control scheme for the virtual power plant; for the GA population particles, perform selection, crossover, and mutation, and evaluate the fitness of each GA population particle according to the specific value of the objective function F; among the GA population particles, the smaller the objective function F of a specific particle, the higher the fitness of the particle, and the more likely it is to be selected into the next generation. Calculate the probability P of each particle being selected into the next generation by operating on the objective function F of each particle in the GA population particles; and perform a crossover operation on all the particles in the GA population particles. Extract a preset number of particles according to the probability P of each particle being selected into the next generation. Randomly exchange the specific values of each dimension in the multi-dimensional vectors corresponding to the selected particles through the crossover operation to simulate partial gene exchange and obtain the new solutions after the crossover operation. Convert the new solutions obtained after the crossover operation into PSO population particles and initialize them to generate a set of initial PSO population particles, where each particle represents a control scheme for the virtual power plant. Input all the PSO population particles obtained by initializing the new solutions after the crossover operation into the PSO velocity update and position update formulas to perform PSO population optimization.

[0059] Among them, the GA population particles and PSO population particles are both multi-dimensional vectors, and each dimension corresponds to the real-time output power of each power generation device at each moment t, the planned production volume of fossil energy at each moment t, the real-time output power of the power generation device at each moment t, the transmission scheduling symbol C(i1, i2) of all upstream devices i1 and downstream devices i2 at each moment t, and the power transmission power between any upstream device i1 and downstream device i2 at each moment t , the total input power and total output power of the energy storage device at each moment t, and the power consumption limit power at the load end at each moment t.

[0060] The PSO velocity update and position update formulas are as follows: ; where is the update formula for the particle velocity, where is the velocity of particle k in the (q + 1)-th operation after update; where k is the particle number index and q is the operation iteration number index; where w is the preset inertia weight, which controls the influence degree of the particle's historical velocity on the velocity after update in the next operation, where is the velocity of particle k in the q-th operation; where c1 and c2 are the preset learning factors, which respectively control the dependence degrees of the particle on the personal best position and the global best position; where r1 and r2 are random numbers between [0, 1]; where is the position of particle k in the q-th operation, which represents the multi-dimensional vector of the virtual power plant control scheme; among them is the personal best solution of particle k, that is, the particle position where the objective function F is minimized during the iterative operation of particle k; among them is the global best solution, that is, the particle position where the objective function F is minimized during the iterative operation of all particles; among them is the particle position update formula.

[0061] Let the stop conditions for the iterative operation of PSO population optimization be: Condition 1: The iteration number q is greater than the preset upper limit qmax of the optimization operation times; Condition 2: The updated velocities of all particles k are all convergent;

[0062] It should be noted that GA and PSO are two common heuristic optimization algorithms. The genetic algorithm is an optimization algorithm that simulates natural selection and genetic principles. It gradually approaches the optimal solution by simulating the evolution process of biological species. The core ideas of GA include selection, crossover, and mutation; Particle Swarm Optimization is a swarm intelligence optimization algorithm that simulates the foraging behavior of bird flocks. Each solution is regarded as a particle, and the particle updates its position according to its own experience and the experience of its neighbors to find the optimal solution to the problem.

[0063] Obtain the multi-dimensional vector corresponding to the global best solution, and match the parameter explanations corresponding to each dimension from it to obtain the virtual power plant optimization control scheme.

[0064] The virtual power plant optimization control scheme described above includes: The scheduling strategy of power generation equipment, that is, the real-time output power of power generation equipment at each moment t and the planned production volume of fossil energy at each moment t; The scheduling strategy of the transmission network, that is, the transmission dispatch symbol at each moment t; The scheduling strategy of energy storage equipment, that is, the total input power and total output power of energy storage equipment at each moment t.

[0065] The scheduling strategy of the load side, that is, the power limit of the load side electricity consumption at each moment t.

[0066] Step 4: Cooperative optimization signal matching; Furthermore, based on the virtual power plant optimization control scheme, match the corresponding control signals and save them to Layer 3: Scheduling Optimization Layer. The control signals include: Power regulation signal at the power generation end: The signal for increasing / decreasing the production volume of fossil energy at each moment t, and the target quality of increasing / decreasing production; Power storage terminal control signal: the charge and discharge instructions of the power storage terminal at each moment t, and the target power of the power storage terminal for charging / discharging; Power transmission terminal control signal: the turn-on / turn-off instructions of the power transmission terminal equipment at each moment t; Load terminal load regulation signal: the power limit instruction of the load terminal for grid dispatching at each moment t, and the target power of the restricted load terminal.

[0067] Send the control signals at each moment t saved in the third layer: the scheduling optimization layer to the management platform for manual review. After obtaining the confirmation of the manual review, output to the first layer: the real layer for the execution of the control signals.

[0068] Step Five: Execution and supervision of the optimized control strategy; During the execution of the control signals, the execution effect is jointly supervised through the PID supervisor and the objective function. The specific process is as follows: Obtain the target quality of the increase / decrease in the production of fossil energy contained in the power generation terminal power regulation signal; calculate the difference between the target quality of the increase / decrease and the actual increase / decrease amount at every preset time interval t to obtain the first execution deviation.

[0069] Obtain the target power of charging / discharging contained in the power storage terminal control signal; calculate the difference between the target power of charging / discharging and the actual charging / discharging power at every preset time interval t to obtain the second execution deviation.

[0070] The target power of the load terminal contained in the load terminal load regulation signal; calculate the difference between the target power of the load terminal and the actual load terminal power at every preset time interval t to obtain the third execution deviation.

[0071] Take the average value of the first, second, and third execution deviations as the total execution deviation e(t) and input it into the PID supervisor: Calculate the PID deviation eigenvalue u(t); where are the preset weight coefficients of the preset proportional, integral, and differential respectively; where t1 is the generation time of the control signal, that is, the time when the control instruction is output after the operation after receiving the optimization operation instruction sent by the administrator for the last time.

[0072] When it is recognized that the PID deviation eigenvalue u(t) is greater than the first preset threshold, or it is recognized that the increasing speed of the final optimization objective function F with respect to time t is greater than the second preset threshold, it is determined that there is a huge deviation between the control signal and the actual control effect, and the isolation between the first layer: the real layer and the third layer: the scheduling optimization layer is executed, and the generation of all new control signals and the execution of the existing control signals are stopped.

[0073] It should be understood that the terms "comprising" and "including" as used in the specification and claims of this disclosure indicate the presence of the described features, wholes, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their combinations.

[0074] It should also be understood that the terminology used herein in the specification of this disclosure is for the purpose of describing particular embodiments only and is not intended to limit the disclosure. As used in the specification and claims of this disclosure, unless the context clearly dictates otherwise, the singular forms "a," "an," and "the" are intended to include the plural forms. It should be further understood that the term "and / or" as used in the specification and claims of this disclosure refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations; The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the present invention to the specific embodiments described. Obviously, many modifications and variations can be made according to the content of this specification. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the present invention, so that those skilled in the art can well understand and utilize the present invention. The present invention is only limited by the claims and their full scope and equivalents.

Claims

1. A multi-level collaborative optimization control method for virtual power plants based on digital twins, characterized in that: The following steps are involved: Step 1: Data collection and digital twin model construction; The operation data of each device in the virtual power plant from the power generation end, the power transmission end, the power storage end to the load end are collected in real time through the sensor group and the information collection equipment; the environmental data is collected in real time from the environmental data sensor; a real-time simulation model of the virtual power plant is established through the digital twin technology, including the physical entity, the operation status, the operation data, the environmental data, the energy flow and the collaborative relationship between the devices; the collected operation data and environmental data are input into the real-time simulation model; Step 2: Establish a multi-level collaborative optimization model; In the real-time simulation model of the virtual power plant, a multi-level classification is performed to obtain level 1: reality layer; level 2: virtual layer and level 3: scheduling optimization layer; in level 2: virtual layer, a multi-objective optimization function for scheduling optimization operation is set; Step 3: Data processing and collaborative optimization; Based on the joint optimization model of weight coefficient allocation, GA genetic algorithm and PSO particle swarm optimization, multi-objective optimization calculation is carried out to obtain the optimal control scheme of the virtual power plant; Step 4: Collaboratively optimize signal matching; Match the corresponding control signal based on the virtual power plant optimization control scheme and save it to level three: scheduling optimization layer; send it to the management platform for manual review, and after the manual review is confirmed, output it to level one: reality layer for control signal execution; Step 5: Optimize control strategy execution and supervision; During the execution of the control signal, the execution effect is jointly supervised by the PID supervisor and the objective function.

2. The multi-level collaborative optimization control method for virtual power plants based on digital twins according to claim 1 is characterized in that: The specific process of multi-level classification in the real-time simulation model of the virtual power plant is as follows: Level 1: Reality layer, including power generation equipment, transmission equipment, energy storage equipment and load-end equipment in each power generation end, transmission end, storage end and load end, as well as the collected operation data and environmental data of each device. The input data of level 1 is the real data set ; Level 2: Virtual layer, including the power generation prediction model at the power generation end, the topological structure model at the power transmission end, the scheduling and allocation model at the power storage end, and the load prediction model at the load end; Level three: Scheduling optimization layer, including several scheduling optimization signals generated by the superposition and comparison of the real layer and the virtual layer.

3. The multi-level collaborative optimization control method of a virtual power plant based on digital twin according to claim 2 is characterized in that: The power generation prediction model at the power generation end is as follows: Power generation prediction model: including fossil energy power generation prediction sub-model, photovoltaic power generation prediction sub-model and wind power generation prediction sub-model. The mathematical model formula is: ;in, is the predicted total output power of traditional fossil energy power generation equipment at time t; is the predicted value of the total output power of the photovoltaic power generation equipment at the future time t; is the predicted value of the total output power of the wind power generation equipment at the future time t; is the predicted value of total power generation at future time t; and are support vector regression functions trained for traditional fossil energy power generation equipment, photovoltaic power generation equipment, and wind power generation equipment, respectively, and are used to predict the power generation of traditional fossil energy, photovoltaic, and wind power at time t in the future; is the planned output of fossil energy at time t, is the total output power of traditional fossil energy power generation equipment at the current moment, is the predicted average temperature of the power generation equipment at the future time t; is the predicted value of light intensity at the future time t, is the total output power of the photovoltaic power generation equipment at the current moment; is the predicted wind speed at future time t, is the total output power of the wind power generation equipment at the current moment; and are the prediction errors of the fossil energy power generation prediction sub-model, photovoltaic power generation prediction sub-model and wind power generation prediction sub-model, which obey Gaussian distribution; It is the predicted value of the total output power of all devices at the power generation end at the future time t.

4. The multi-level collaborative optimization control method for virtual power plants based on digital twins according to claim 2 is characterized in that: The topological structure and transmission loss model of the transmission end are as follows: Where i1 and i2 are the ID of the upstream device and the ID of the downstream device connected to each transmission device, respectively. I is a set of IDs of all power generation equipment, energy storage equipment and load-end equipment. C (i1, i2) is the transmission dispatch symbol. The value of C (i1, i2) is 1, which means that the transmission work is carried out between the upstream device i1 and the downstream device i2. The value of C (i1, i2) is 0, which means that the transmission work is not carried out between the upstream device i1 and the downstream device i2. represents the transmission power between the upstream device i1 and the downstream device i2, where V(i1, i2) is the transmission line voltage between the upstream device i1 and the downstream device i2.

5. The multi-level collaborative optimization control method for virtual power plants based on digital twins according to claim 2 is characterized in that: The dispatching and allocation model of the power storage terminal is as follows: ;in is the predicted storage capacity of all energy storage devices at time t, where is the actual storage capacity of the energy storage device at the previous historical moment t; is the total input power and total output power of all energy storage devices collected at time x; are the charge and discharge efficiencies respectively; Forecast management costs for charging and discharging behavior, are the management costs generated by the charging and discharging scheduling of unit power respectively; and They are the maximum threshold of total input power, the maximum threshold of total output power and the maximum threshold of storage capacity of all energy storage devices respectively.

6. The multi-level collaborative optimization control method for virtual power plants based on digital twins according to claim 1 is characterized in that: The multi-objective optimization function includes: Minimize energy loss objective optimization function: , that is, the total power loss at the transmission end is minimized; Balance supply and demand objective optimization function: , that is, the total output power of the power generation end and the power storage end is equal to the total output power of the load end; Minimize the cost of electricity generation objective optimization function: , that is, to minimize the cost of fossil energy production and storage dispatch management, among which The cost required to put a unit of fossil energy into production; Minimum carbon emission objective optimization function: , that is, the planned output of fossil energy at the power generation end is minimized.

7. The multi-level collaborative optimization control method for virtual power plants based on digital twins according to claim 1 is characterized in that: The specific process of obtaining the optimal control scheme of the virtual power plant through calculation is as follows: The weight coefficients λ1, λ2, λ3 and λ4 are respectively assigned to the objective optimization function f1 of minimizing energy loss, the objective optimization function f2 of balancing supply and demand, the objective optimization function f3 of minimizing power generation cost and the objective optimization function f4 of minimizing carbon emission, which respectively represent the importance of each multi-objective optimization function; the sum of λ1, λ2, λ3 and λ4 is 1, and the specific values ​​of each are dynamically adjusted according to the scheduling requirements; the objective optimization function f1 of minimizing energy loss, the objective optimization function f2 of balancing supply and demand, the objective optimization function f3 of minimizing power generation cost and the objective optimization function f4 of minimizing carbon emission are weighted and summed according to the weight coefficients assigned to each function, and the final optimization objective function F is obtained; Whenever an optimization operation instruction is received from the administrator, a joint optimization model based on GA genetic algorithm and PSO particle swarm optimization is built to perform multi-objective optimization operations, and the final optimization objective function is solved. Through selection, crossover and mutation operations, the virtual power plant optimization control scheme that satisfies the conditions is obtained: the final optimization objective function F is minimized; The virtual power plant optimization control scheme includes: The dispatching strategy of power generation equipment, i.e. the real-time output power of power generation equipment at each time t and the planned output of fossil energy at each time t; The dispatching strategy of the transmission network, i.e. the transmission dispatch symbol at each time t; The scheduling strategy of the power storage device, that is, the total input power and total output power of the power storage device at each time t; The scheduling strategy of the load end, that is, the power limit of the load end power consumption at each time t.

8. The multi-level collaborative optimization control method for virtual power plants based on digital twins according to claim 1 is characterized in that: The control signal includes: Power regulation signal at the power generation end: the increase / decrease signal of fossil energy production at each time t, as well as the target quality of increase / decrease; Power storage end control signal: the charge and discharge instructions of the power storage end at each time t, and the target power of charging / discharging of the power storage end; Transmission end control signal: the on / off instruction of the transmission end equipment at each time t; Load regulation signal at the load end: the power limit instruction at the load end dispatched by the power grid at each time t, and the restricted load end target power.

9. The multi-level collaborative optimization control method for virtual power plants based on digital twins according to claim 1 is characterized in that: Based on the joint optimization model of weight coefficient allocation, GA genetic algorithm and PSO particle swarm optimization, multi-objective optimization calculation is performed. The specific process of obtaining the optimal control scheme of the virtual power plant through calculation is as follows: Assign weight coefficients λ1, λ2, λ3 and λ4 to the objective optimization function of minimizing energy loss, the objective optimization function of balancing supply and demand, the objective optimization function of minimizing power generation cost and the objective optimization function of minimizing carbon emission, respectively, which represent the importance of each multi-objective optimization function; the sum of λ1, λ2, λ3 and λ4 is 1, and their specific values ​​are dynamically adjusted according to the scheduling requirements; perform weighted summation of the objective optimization function of minimizing energy loss, the objective optimization function of balancing supply and demand, the objective optimization function of minimizing power generation cost and the objective optimization function of minimizing carbon emission according to their assigned weight coefficients, and obtain the final optimization objective function F; The optimization objective function F is specifically: ; Furthermore, whenever an optimization operation instruction is received from the administrator, a joint optimization model based on GA genetic algorithm and PSO particle swarm optimization is built to perform multi-objective optimization operations, and the final optimization objective function F is solved. Through selection, crossover and mutation operations, the virtual power plant optimization control scheme that satisfies the condition: the final optimization objective function F is minimized is obtained. The specific process is: Initialize the population particles of GA and PSO, initialize the GA population particles, generate a group of GA population particles representing the initial candidate solutions, each candidate solution represents a control scheme of the virtual power plant; each particle represents a control scheme of the virtual power plant; select, crossover and mutate the GA population particles, and evaluate the fitness of each GA population particle according to the specific value of the objective function F; among the GA population particles, the smaller the objective function F of a specific particle is, the higher the fitness of the particle is, and the more likely it is to be selected to enter the next generation. The probability P of each particle being selected to enter the next generation is obtained by calculating the objective function F of each particle in the GA population particles; and perform a crossover operation on all particles in the GA population particles, extract a preset number of particles with the probability P of each particle being selected to enter the next generation, and randomly exchange the specific values ​​of each dimension in the multidimensional vector corresponding to the extracted particle through the crossover operation, and simulate partial gene exchange to obtain a new solution after the crossover operation; The new solution obtained after the crossover operation is converted into PSO population particles, initialized, and a set of initial PSO population particles is generated, where each particle represents a control scheme of the virtual power plant; All PSO population particles obtained by initializing the new solution after the crossover operation are input into the PSO velocity update and position update formulas to perform PSO population optimization; Among them, the GA population particles and PSO population particles are multidimensional vectors, each dimension corresponds to the real-time output power of each power generation equipment at each time t, the planned output of fossil energy at each time t, the real-time output power of the power generation equipment at each time t, the transmission dispatch symbol C (i1, i2) of all upstream devices i1 and downstream devices i2 at each time t, and the transmission power of any upstream device i1 and downstream device i2 at each time t. , the total input power and total output power of the power storage device at each time t, and the power limit of the load end at each time t; The PSO velocity update and position update formulas are: ;in is the update formula of particle velocity, where is the speed of particle k in the q+1th operation after the update; k is the particle number index, q is the operation iteration number index; w is the preset inertia weight, which controls the influence of the particle's historical speed on the updated speed in the next operation, The speed of particle k in the qth operation; c1 and c2 are preset learning factors, which control the degree of dependence of the particle on the personal best position and the global best position respectively; r1 and r2 are random numbers between [0, 1]; is the position of particle k in the qth operation, that is, a multidimensional vector representing the virtual power plant control scheme; is the personal optimal solution of particle k, that is, the particle position of particle k that minimizes the objective function F in the iterative operation; is the global optimal solution, that is, the particle position where all particles optimize the minimum objective function F in the iterative operation; Update formula for particle position; The conditions for stopping the iteration operation of PSO population optimization are: Condition 1: The number of iterations q is greater than the preset upper limit of the number of optimization operations qmax; Condition 2: The updated speed of all particles k All converge; When it is identified that condition 1 or condition 2 is met, the optimization is determined to be successful, the iterative operation of the PSO velocity update and position update formulas is stopped, and the global optimal solution is output as the final optimization result; Obtain the multidimensional vector corresponding to the global optimal solution, and match the parameter interpretation corresponding to each dimension to obtain the optimal control solution for the virtual power plant.

10. The multi-level collaborative optimization control method of a virtual power plant based on digital twin according to claim 1, characterized in that: The specific process of joint supervision of execution effect through PID supervisor and objective function is as follows: Obtaining the target quality of the increase / decrease in the output of fossil energy contained in the power regulation signal of the power generation end; calculating the difference between the target quality of the increase / decrease in output and the actual increase / decrease in output at every preset time interval t, and obtaining a first execution deviation; Obtaining the target power of charging / discharging contained in the control signal of the power storage end; calculating the difference between the target power of charging / discharging and the actual charging / discharging power at every preset time interval t to obtain a second execution deviation; The load end target power included in the load end load regulation signal; every preset time interval t, calculating the difference between the load end target power and the actual load end power to obtain a third execution deviation; The average of the first, second and third execution deviations is input into the PID supervisor as the total execution deviation e(t): Calculate the PID deviation characteristic value u(t); where are preset weight coefficients for the preset proportion, integral and differential respectively; Where t1 is the time when the control signal is generated, that is, the time when the control instruction is output after the optimization operation instruction is received from the administrator most recently; When it is identified that the PID deviation characteristic value u(t) is greater than the first preset threshold, or it is identified that the increase rate of the final optimization objective function F over time t is greater than the second preset threshold, it is determined that there is a huge deviation between the control signal and the actual control effect, and the isolation between the execution level one: the reality layer and the level three: the scheduling optimization layer is stopped, and the generation of all new control signals and the execution of existing control signals are stopped.

Images