A virtual power plant energy optimization scheduling method and scheduling system
Through the energy optimization scheduling method and scheduling system of virtual power plant, combined with cogeneration, wind power generation, photovoltaic power generation and demand response models, the scheduling simulation is used to solve the problem of lack of top-level design and algorithm guidance in virtual power plant technology. It improves the robustness of the system and the efficiency of information integration, and expands the expansion of the system.
Patent Information
- Application Number
- CN202410649444.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-24
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-05-24
AI Technical Summary
Virtual power plant technology lacks top-level design and algorithm guidance, management and supervision are unclear, industry standards are not unified, and all parties are not willing to participate.
Design a virtual power plant energy optimization scheduling method and scheduling system, including operator subsystem, scheduling subsystem, user subsystem and multi-virtual power plant joint module, adopting cogeneration, wind power generation, photovoltaic power generation, load monitoring and demand response models , combine convolutional neural network for scheduling simulation and optimization.
Improve the top-level design of virtual power plants, provide algorithm guidance for operation and scheduling, improve data credibility and system robustness, alleviate information island problems, improve information integration efficiency and data collaboration efficiency, and expand the system expansion.
Smart Images

Figure CN118428686B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of virtual power plants, and in particular to a virtual power plant energy optimization scheduling method and scheduling system. Background Art
[0002] A virtual power plant is a power coordination management system that uses advanced information and communication technologies and software systems to achieve the aggregation and coordinated optimization of distributed power sources (such as renewable energy generation and energy storage systems) and controllable loads, so as to participate in the power market and power grid operation as a special power plant. A virtual power plant is not a real power plant, but a smart grid technology that uses a distributed power management system to participate in the operation and dispatch of the power grid to achieve "source-load-grid" aggregation optimization. The grid is balanced by coordinating power generation resources and adjusting some power demand, reducing peak power consumption and increasing "power elasticity". The operation platform of the virtual power plant connects the power trading center and resource aggregators, organizes resource entities to participate in various power market transactions, and completes relevant settlements and profit distribution. Therefore, the design and innovation of the pricing mechanism and profit distribution mechanism of the virtual power plant operation platform for different resource combinations is the key to stimulating various user-side resources to actively participate in the virtual power plant and improve the overall benefits of the virtual power plant. The virtual power plant has many advantages such as resource integration optimization, flexibility and response speed, cost-effectiveness, environmental protection and sustainability. It can be applied to urban energy systems, microgrids and new energy fields to reduce operating costs and improve energy utilization efficiency by optimizing energy distribution and utilization. With the transformation of energy structure and the reform of power market, the status and role of virtual power plants in the power system will become increasingly important.
[0003] However, virtual power plant technology also faces some challenges and risks, such as lack of top-level design and algorithm guidance, unclear management and supervision, inconsistent industry standards, and weak willingness of all parties to participate. Summary of the invention
[0004] In order to overcome the problem of lack of appropriate top-level design and algorithm guidance faced by virtual power plant technology in the background technology, an embodiment of the present invention provides a virtual power plant energy optimization scheduling method and scheduling system.
[0005] A virtual power plant energy optimization scheduling method and scheduling system, including an operator subsystem, a scheduling subsystem, a user subsystem and a multi-virtual power plant joint module, wherein the operator subsystem includes a fossil energy module and a renewable energy module, the scheduling subsystem includes a prediction module, a scheduling simulation module and a virtual power plant scheduling center, and the user subsystem includes a load monitoring module and a demand response load module.
[0006] As a preferred embodiment of the present invention, the fossil energy module adopts a cogeneration energy production method, and uses the electric energy generated by natural gas in the power turbine as electric power output; and uses the steam generated by coal in the thermal boiler as thermal output. The turbine waste heat generated by the power turbine is divided into turbine waste heat 1 and turbine waste heat 2, and turbine waste heat 1 is output to the low-temperature waste heat power generation device, and turbine waste heat 2 is output to the waste heat boiler. The low-temperature waste heat power generation device realizes heat-to-electricity conversion through organic Rankine cycle technology, converts turbine waste heat 1 into electric energy, and incorporates it into the power output line of the power turbine; the waste heat boiler realizes waste heat reuse through heat energy recovery, converts turbine waste heat 2 into steam, and incorporates it into the thermal output line of the thermal boiler.
[0007] Get the power generation efficiency η of the power turbine converting natural gas into electricity 轮机电 and the thermal efficiency η of the generator turbine converting natural gas into thermal energy 轮机热 ; Obtain the thermal efficiency η of the thermal boiler converting coal into thermal energy 锅炉 ; Obtain the efficiency η of low-temperature waste heat power generation device in converting heat energy into electrical energy 余热-电 ; Obtain the efficiency η of heat recovery from waste heat boiler 余热-热 ; Obtain the standard calorific value f of natural gas power generation per unit volume 天然气 ; Get the standard value of heat production per unit mass of coal f 煤 .
[0008] Establish a dispatch model for the fossil energy module:
[0009]
[0010] where λ 天然气 is the volume of natural gas consumed per unit time; 煤 is the mass of coal consumed per unit time; P 电输出 is the total output power of the power output line; H 热输出 is the total output power of the thermal output circuit; Q 1 is the output power of turbine waste heat; Q 1 is the output power of turbine waste heat 2; α is the distribution coefficient of turbine waste heat 1 and turbine waste heat 2.
[0011] As a preferred embodiment of the present invention, the clean energy module includes an energy production method of wind power generation and photovoltaic power generation, and establishes a Welbull probability distribution model of wind speed: Where v is the wind speed; c is the scale parameter; k is the shape parameter. Wind speed data is obtained through wind speed sensors arranged on each wind turbine at preset intervals. All wind speed data are recorded and fitted with the Welb probability distribution model. The scale parameter c and shape parameter k are determined by the maximum likelihood estimation method. A power generation model for a single wind turbine is established: Where i is the fan number; v is the wind speed; vmin is the minimum design wind speed; v max is the maximum design wind speed; P 风电0 is the rated power of the wind turbine; P 风电 is the actual power generation of the wind turbine. Get the number of all wind turbines in the area n1, through the formula Calculate the total electricity generated by wind power.
[0012] The power generation of solar panels in the photovoltaic power generation method is affected by the light intensity and is linearly related to the light intensity. A beta probability model of light intensity is established: Where r is the light intensity; rmax is the maximum light intensity; Γ(x) is the gamma function, expressed as The gamma function in the beta probability model expression of light intensity is: α is shape parameter 1; β is shape parameter 2. Get the maximum historical light intensity rmax in the area where the solar panel is located in a year, obtain the light intensity r through the light sensor at preset time intervals, record all light intensity data and fit it with the beta probability model of light intensity, and determine the values of shape parameter 1 α and shape parameter 2 β by maximum likelihood estimation method. Get the energy conversion efficiency and area of each solar panel, and establish the power generation model of a single solar panel: p 光电,j =η 光电,j ×S j ×r, where j is the solar panel number; p 光电,j is the power generated by the jth solar panel; S j is the area of the jth solar panel; r is the light intensity. To obtain the number of all solar panels n2, use the formula Calculate the total electricity generated by photovoltaic power generation.
[0013] As a preferred method of the present invention, the load monitoring module obtains the power readings of the smart meters of each user of the virtual power plant at preset intervals, calculates the sum of all power readings to obtain the load power Pload at that time, generates a vector (time t, load power Pload), uses time as the horizontal coordinate and load power as the vertical coordinate to generate a two-dimensional rectangular coordinate system, fills all points (time t, load power Pload) into the coordinate system, and then connects all points with a smooth straight line to generate a curve of load power changing with time, which is recorded as a daily load curve. In the daily load curve, a peak load P is set. 高峰 and valley load P 低谷 The part of the load power that is higher than the peak load is recorded as the peak load, the part of the load power that is between the peak load and the valley load is recorded as the level load, and the part of the load power that is less than the valley load is recorded as the valley load.
[0014] As a preferred embodiment of the present invention, the demand response module establishes a price-based demand response model and an alternative demand response model. The price-based demand response model adjusts the electricity price to a preset peak electricity price C during the peak electricity consumption period. 高峰 ; During the original time period of the electricity consumption level, the electricity price is adjusted to the preset plain electricity price C 平原 ; During the off-peak period, adjust the electricity price to the preset off-peak price C 低谷 The area of the peak power consumption is extracted as the peak power consumption E 高峰 , extract the area of the original part of the level as the plain charge E 平原 , extract the area of the valley part as the valley electricity E 低谷 The peak electricity price C is calculated every preset time t. 高峰 、Plain electricity price C 平原 , Off-peak electricity price C 低谷 , Peak power consumption E 高峰 、Plain electricity E 平原 and valley power E 低谷 . Every preset time, count the peak electricity price adjustment ΔC compared with the last electricity price change. 高峰 , Plain electricity price adjustment ΔC 平原 , Off-peak electricity price adjustment ΔC 低谷 , Peak power change value ΔE 高峰 , Plain electricity change value ΔE 平原 And the valley power change value ΔE 低谷 . A price-based demand response model is established to describe this behavior: Where E is the response compliance matrix, Where eij is the coefficient of the obedience matrix, and n is the number of recorded data sets. The approximate solution of the response obedience matrix E is obtained based on the vector autoregression model (VAR) every preset time T, and the preset time T is greater than the preset time t. By formula E 日总 =E 高峰 +E 平原 +E 低谷 Calculate the total daily load.
[0015] Query the unit energy standard calorific value of electricity Qelectricity, the unit energy standard calorific value of steam Qheat and the unit energy standard calorific value of natural gas Qgas. Establish an alternative demand response model Where A is the replaced energy, A = electricity, heat, gas; B is the replacement energy, B = electricity, heat, gas; ΔD A is the load change of the replaced energy source A; ΔD B is the load change of alternative energy B; AB is the conversion coefficient of replaced energy A to alternative energy B; ηA is the utilization efficiency of replaced energy A; ηB is the utilization efficiency of alternative energy B.
[0016] The constraints of the alternative demand response model are: Where ΔD A min is the minimum value of the replaced energy A; ΔD A max is the maximum value of the replaced energy A; ΔD B min is the minimum value of alternative energy B; ΔD B max is the maximum value of alternative energy B.
[0017] As a preferred embodiment of the present invention, the prediction module establishes a maximum revenue objective function of the virtual power plant:
[0018]
[0019]
[0020] PRO 电 Profits from selling electricity to users; PRO 热 Profits from selling heating energy to users; PRO 热 The profit from selling natural gas to users; COST 热 The selling price of heating per unit power, COST 气 COST is the price of natural gas per unit power; 煤,气 is the purchase cost of coal and natural gas, C 煤 is the purchase price of coal per unit mass, C 天然气 is the purchase price of natural gas per unit volume; COST 维护 is the maintenance cost of wind turbines and solar panels, λ1 is the standard operation and maintenance cost of a single wind turbine, and λ2 is the standard operation and maintenance cost of a single solar panel; COST 购入 is the spot price of electricity purchased from the power grid. 购入 When COST is positive, it represents the spot price cost of purchasing electricity from the power grid. 购入 Negative means the spot price income from selling excess electricity to the public grid; C 电 It is the purchase price of spot electricity per unit energy.
[0021] The dispatch model of fossil energy modules, the power generation model of a single wind turbine, the power generation model of a single solar panel, the price-based demand response model and the alternative demand response model are comprehensively considered to solve the objective function, and the energy selling price is constrained by the following constraints:
[0022] Selling price constraints: The peak electricity price is higher than the plain electricity price, the plain electricity price is higher than the valley electricity price, the peak electricity price is lower than the peak electricity price upper limit, and the valley electricity price is lower than the minimum electricity price lower limit; the selling price of heating per unit power is higher than the minimum value and lower than the maximum value; the selling price of natural gas per unit power is higher than the minimum value and lower than the maximum value;
[0023] Solve the maximum profit objective function of the virtual power plant under the constraints, and record the natural gas volume consumed per unit time, the quality of coal consumed per unit time, and the peak electricity price C in the maximum profit solution. 高峰 、Plain electricity price C 平原 and valley electricity price C 低谷 , recorded as the optimal scheduling solution;
[0024] As a preferred embodiment of the present invention, the scheduling simulation module establishes a convolutional neural network to simulate the optimal scheduling solution.
[0025] A four-layer convolutional neural network model is established, where f(x) and g(x) are dynamic convolution kernels, and λ A,B , B,C and λ C,D is a dynamic weight factor. X1, X2, ..., Xa are input values, and Y1, Y2 and Y3 are output values.
[0026] The input values of the four-layer convolutional neural network X1, X2, ..., Xa are a conditional matrix; each conditional matrix includes a benefit sub-matrix, a cost sub-matrix, an energy layout sub-matrix and a response compliance sub-matrix. The benefit sub-matrix includes the peak electricity price C 高峰 、Plain electricity price C 平原 , Off-peak electricity price C 低谷 、The selling price of heating per unit power COST 热 、COST of natural gas per unit power 气 and the purchase price of spot electricity per unit energy C 电 ; The cost sub-matrix includes the purchase price of unit mass coal C 煤 , the purchase price of natural gas per unit volume C 天然气 , the standard operation and maintenance cost of a single wind turbine λ1, the standard operation and maintenance cost of a single solar panel λ2; the energy layout sub-matrix includes the coal mass consumed per unit time in the fossil energy module λ 煤 , the volume of natural gas consumed per unit time in the fossil energy module λ 天然气 , the number of wind turbines i and the number of solar panels j; the response obedience submatrix comes from the response obedience matrix E1 in the price-type demand response model, and the output values include the virtual power plant revenue Y1, the daily total load Y2, and the spot electricity price cost Y3 purchased from the power grid.
[0027] Convolutional layer 1 and convolutional layer 2 perform data pooling after completing the convolution calculation to extract matrix features.
[0028] The training steps of the four-layer convolutional neural network are as follows:
[0029] S1: Network initialization, initialization of f(x), g(x), λ A,B , B,C and λ C,D .
[0030] S2: Collect data samples, and retrieve data from the operator subsystem and user subsystem at preset intervals to generate a condition matrix and corresponding output values. The condition matrix and the corresponding output values are divided into two categories, which are used for neural network training and comparison of prediction results respectively;
[0031] S3: Calculate the prediction error e
[0032] Where Yd is the actual output value, is the calculated value of the neural network;
[0033] S4: Correct the dynamic weight and convolution kernel, and correct the dynamic weight and convolution kernel by the prediction error e until the prediction error e is less than a preset threshold;
[0034] S5: The process ends when the prediction error e is less than the preset threshold of e, otherwise, the process returns to step S3.
[0035] After the training is completed, the conditional matrix generated by the optimal scheduling solution is substituted for calculation to obtain the result and Obtain the virtual power plant revenue Y1, daily total load Y2, and spot electricity price cost purchased from the power grid in the optimal dispatching plan. Calculate the verification feasibility index θ, where k is the preset influencing factor. When the verification feasibility index is greater than the preset threshold, the simulation verification is considered complete and the optimal scheduling plan is sent to the virtual power plant dispatch center.
[0036] As a preferred embodiment of the present invention, the virtual power plant dispatching center obtains the peak electricity price C in the optimal dispatching scheme after simulation verification. 高峰 、Plain electricity price C 平原 , Off-peak electricity price C 低谷 、The selling price of heating per unit power COST 热 、COST of natural gas per unit power 气 , coal mass consumed per unit time λ 煤 , the volume of natural gas consumed per unit time λ 天然气 , the number of wind turbines i and the number of solar panels j. Generate the target data vector A = (C 高峰 , C 平原 , C 低谷 , COST 热 , COST气 ,λ 煤 ,λ 天然气 , i, j).
[0037] Get current dispatch data, including current peak electricity prices Current Plain Electricity Price Current low electricity prices Current selling price of heating per unit power Current price of natural gas per unit power The coal mass consumed per unit time at present The volume of natural gas consumed per unit time The current number of wind turbines i0 and solar panels j0 generates the status data vector
[0038] By formula Calculate a single adjustment value where ΔX is ΔC 高峰 , ΔC 平原 , ΔC 低谷 , ΔCOST 热 , ΔCOST 气 , Δλ 煤 , Δλ 天然气 , Δi, Δj, represent single adjustment values, X is the element in the target data vector A; X 0 is the element in the current data vector; ε is the preset optimization speed factor. Generate adjustment data vector (ΔC 高峰 , ΔC 平原 , ΔC 低谷 , ΔCOST 热 , ΔCOST 气 , Δλ 煤 , Δλ 天然气 , Δi, Δj). Each optimization adjustment increases or decreases the corresponding value according to the value of the element in the adjustment data vector, and a total of ε optimization adjustments are performed.
[0039] As a preferred embodiment of the present invention, a multi-virtual power plant joint module integrates multiple virtual power plants into the public power grid to jointly play a scheduling optimization role, output excess power to the public power grid, and purchase power from the public power grid when power generation is insufficient.
[0040] A virtual power plant energy optimization scheduling method comprises the following steps:
[0041] Step 1: Use the energy production method of cogeneration to establish a scheduling model for the fossil energy module;
[0042] Step 2: Establish a Welbull probability distribution model of wind speed and a power generation model of a single wind turbine;
[0043] Step 3: Establish a beta probability model of light intensity and a power generation model of a single solar panel;
[0044] Step 4: Establish a price-based demand response model and an alternative demand response model;
[0045] Step 5: Obtain sensor monitoring data and substitute it into each model to solve the maximum benefit objective function and obtain the optimal scheduling plan;
[0046] Step 6: Establish a convolutional neural network to simulate the optimal scheduling solution;
[0047] Step 7: Generate an adjustment data vector according to the optimal scheduling solution;
[0048] Step 8: Perform scheduling optimization according to the adjusted data vector;
[0049] Step 9: Combine multiple virtual power plants.
[0050] Compared with the prior art, the present invention has the following beneficial effects:
[0051] 1. The present invention simulates the operation mode of a virtual power plant by establishing a fossil energy module scheduling model, a single wind turbine power generation model, a single solar panel power generation model and a demand response model, comprehensively considers the model and actual data to solve the optimal scheduling solution, improves the top-level design of the virtual power plant, and provides an algorithm guidance case for the operation and scheduling of the virtual power plant;
[0052] 2. The present invention solves the optimal scheduling scheme through multi-model joint solution and uses convolutional neural network for verification, which improves the credibility of data, reduces the impact of data disturbance on the entire system, improves the robustness of the system, alleviates the problem of information islands, and improves the information integration efficiency and data coordination efficiency of the entire system;
[0053] 3. The present invention constructs a clean energy ecological network in which multiple virtual power plants are integrated into the public power grid through a multi-virtual power plant joint module, thereby expanding the scalability of the entire system and completing the three-dimensional energy optimization and scheduling of virtual power plants. BRIEF DESCRIPTION OF THE DRAWINGS
[0054] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.
[0055] Figure 1 is a system block diagram of the present invention;
[0056] Figure 2 A schematic diagram of the energy production method for combined heat and power;
[0057] Figure 3 It is a schematic diagram of daily load curve;
[0058] Figure 4 Schematic diagram of the topological structure of the convolutional neural network;
[0059] Figure 5 It is a schematic diagram of the topological structure of the conditional matrix;
[0060] Figure 6 Schematic diagram of the topological structure of the pool;
[0061] Figure 7 This is a schematic diagram of the combination of multiple virtual power plants;
[0062] Figure 8 The present invention is a flow chart of the method. DETAILED DESCRIPTION
[0063] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0064] See also Figure 1 As shown, a virtual power plant energy optimization scheduling method and scheduling system include an operator subsystem, a scheduling subsystem, a user subsystem and a multi-virtual power plant joint module, wherein the operator subsystem includes a fossil energy module and a renewable energy module, the scheduling subsystem includes a prediction module, a scheduling simulation module and a virtual power plant scheduling center, and the user subsystem includes a load monitoring module and a demand response load module.
[0065] See also Figure 2 As shown, the fossil energy module adopts the energy production method of cogeneration, and uses the electricity generated by natural gas in the power turbine as power output; and uses the steam generated by coal in the thermal boiler as thermal output. The turbine waste heat generated by the power turbine is divided into turbine waste heat 1 and turbine waste heat 2. Turbine waste heat 1 is output to the low-temperature waste heat power generation device, and turbine waste heat 2 is output to the waste heat boiler. The low-temperature waste heat power generation device realizes heat-to-electricity conversion through organic Rankine cycle technology, converts turbine waste heat 1 into electricity, and incorporates it into the power output line of the power turbine; the waste heat boiler realizes waste heat reuse through heat energy recovery, converts turbine waste heat 2 into steam, and incorporates it into the thermal output line of the thermal boiler.
[0066] Get the power generation efficiency η of the power turbine converting natural gas into electricity 轮机电 and the thermal efficiency η of the generator turbine converting natural gas into thermal energy 轮机热 ; Obtain the thermal efficiency η of the thermal boiler converting coal into thermal energy 锅炉 ; Obtain the efficiency η of the low-temperature waste heat power generation device in converting thermal energy into electrical energy 余热-电; Obtain the efficiency η of heat recovery from waste heat boiler 余热-热 ; Obtain the standard calorific value f of natural gas power generation per unit volume 天然气 ; Get the standard value of heat production per unit mass of coal f 煤 .
[0067] Establish a dispatch model for the fossil energy module:
[0068]
[0069] where λ 天然气 is the volume of natural gas consumed per unit time; 煤 is the mass of coal consumed per unit time; P 电输出 is the total output power of the power output line; H 热输出 is the total output power of the thermal output circuit; Q 1 is the output power of turbine waste heat; Q 1 is the output power of turbine waste heat 2; α is the distribution coefficient of turbine waste heat 1 and turbine waste heat 2.
[0070] The clean energy module includes energy production methods of wind power generation and photovoltaic power generation. The wind turbine power generation in the wind power generation method is affected by wind speed, and the wind turbine power generation and wind speed do not strictly conform to a linear relationship. When the wind speed reaches the minimum design wind speed, the wind turbine starts to rotate and output electrical energy; when the wind turbine exceeds the maximum design wind speed, the wind turbine will start the brake device to limit the wind turbine speed. At this time, the power of the wind turbine output electrical energy does not increase with the increase of wind speed. The Welbu probability model can better describe the probability distribution characteristics of wind speed, and establish the Welbu probability distribution model of wind speed: Where v is the wind speed; c is the scale parameter; k is the shape parameter. Wind speed data is obtained through wind speed sensors arranged on each wind turbine at preset intervals. All wind speed data are recorded and fitted with the Welb probability distribution model. The scale parameter c and shape parameter k are determined by the maximum likelihood estimation method. A power generation model for a single wind turbine is established: Where i is the fan number; v is the wind speed; v min is the minimum design wind speed; v max is the maximum design wind speed; P 风电0 is the rated power of the wind turbine; P 风电 is the actual power generation of the wind turbine. Get the number of all wind turbines in the area n1, through the formula Calculate the total electricity generated by wind power.
[0071] The power generation of solar panels in the photovoltaic power generation method is affected by the light intensity and is linearly related to the light intensity. A beta probability model of light intensity is established: Where r is the light intensity; rmax is the maximum light intensity; Γ(x) is the gamma function, expressed as The gamma function in the beta probability model expression of light intensity is: α is shape parameter 1; β is shape parameter 2. Get the maximum historical light intensity rmax in the area where the solar panel is located in a year, obtain the light intensity r through the light sensor at preset time intervals, record all light intensity data and fit it with the beta probability model of light intensity, and determine the values of shape parameter 1 α and shape parameter 2 β by maximum likelihood estimation method. Get the energy conversion efficiency and area of each solar panel, and establish the power generation model of a single solar panel: p 光电,j =η 光电,j ×S j ×r, where j is the solar panel number; p 光电,j is the power generated by the jth solar panel; S j is the area of the jth solar panel; r is the light intensity. To obtain the number of all solar panels n2, use the formula Calculate the total electricity generated by photovoltaic power generation.
[0072] See also Figure 3 As shown, the load monitoring module obtains the power readings of each user's smart meter of the virtual power plant at preset intervals, calculates the sum of all power readings to obtain the load power Pload at that time, generates a vector (time t, load power Pload), uses time as the horizontal coordinate and load power as the vertical coordinate to generate a two-dimensional rectangular coordinate system, fills all points (time t, load power Pload) into the coordinate system, and then connects all points with a smooth straight line to generate a curve of load power changing with time, which is recorded as the daily load curve. In the daily load curve, a peak load P is set 高峰 and valley load P 低谷 The part of the load power that is higher than the peak load is recorded as the peak load, the part of the load power that is between the peak load and the valley load is recorded as the level load, and the part of the load power that is less than the valley load is recorded as the valley load.
[0073] The demand response module establishes a price-based demand response model and an alternative demand response model. The price-based demand response model is designed to simulate the demand response mechanism established by the energy consumption behavior habits of users who turn on electrical appliances during low-peak electricity consumption when facing time-of-use electricity prices; the alternative demand response model is designed to simulate the demand response mechanism established by the energy consumption behavior pattern of users choosing between the three alternative and complementary energy sources of electricity, heat and gas based on price factors.
[0074] The price-based demand response model adjusts the electricity price to the preset peak electricity price C during the peak electricity consumption period. 高峰; During the original time period of the electricity consumption level, the electricity price is adjusted to the preset plain electricity price C 平原 ; During the off-peak period, adjust the electricity price to the preset off-peak price C 低谷 The area of the peak power consumption is extracted as the peak power consumption E 高峰 , extract the area of the original part of the level as the plain charge E 平原 , extract the area of the valley part as the valley electricity E 低谷 In the power market environment, electricity has the attributes of a commodity. Whether users obey price-based demand response and implement peak-shifting electricity consumption is a complex market economic behavior. The peak electricity price C is calculated every preset time t. 高峰 、Plain electricity price C 平原 , Off-peak electricity price C 低谷 , Peak power consumption E 高峰 、Plain electricity E 平原 and valley power E 低谷 . Every preset time, count the peak electricity price adjustment ΔC compared with the last electricity price change. 高峰 , Plain electricity price adjustment ΔC 平原 , Off-peak electricity price adjustment ΔC 低谷 , Peak power change value ΔE 高峰 , the plain electricity change value ΔEplain and the valley electricity change value ΔEvalley. Establish a price-based demand response model to describe this behavior. Where E is the response compliance matrix, Where eij is the coefficient of the obedience matrix, and n is the number of recorded data sets. The approximate solution of the response obedience matrix E is obtained based on the vector autoregression model (VAR) every preset time T, and the preset time T is greater than the preset time t. By formula E 日总 =E 高峰 +E 平原 +E 低谷 Calculate the total daily load.
[0075] Query the unit energy standard calorific value of electricity Qelectricity, the unit energy standard calorific value of steam Qheat and the unit energy standard calorific value of natural gas Qgas. Establish an alternative demand response model Where A is the replaced energy, A = electricity, heat, gas; B is the replacement energy, B = electricity, heat, gas; ΔD A is the load change of the replaced energy source A; ΔD B is the load change of alternative energy B; AB is the conversion coefficient of replaced energy A to alternative energy B; ηA is the utilization efficiency of replaced energy A; ηB is the utilization efficiency of alternative energy B.
[0076] The constraints of the alternative demand response model are: Where ΔD Amin is the minimum value of the replaced energy A; ΔD A max is the maximum value of the replaced energy A; ΔD B min is the minimum value of alternative energy B; ΔD B max is the maximum value of alternative energy B.
[0077] The prediction module sets up the virtual power plant maximum profit objective function:
[0078]
[0079]
[0080] PRO 电 Profits from selling electricity to users; PRO 热 Profits from selling heating energy to users; PRO 热 The profit from selling natural gas to users; COST 热 The selling price of heating per unit power, COST 气 COST is the price of natural gas per unit power; 煤,气 is the purchase cost of coal and natural gas, C 煤 is the purchase price of coal per unit mass, C 天然气 is the purchase price of natural gas per unit volume; COST 维护 is the maintenance cost of wind turbines and solar panels, λ1 is the standard operation and maintenance cost of a single wind turbine, and λ2 is the standard operation and maintenance cost of a single solar panel; COST 购入 is the spot price of electricity purchased from the power grid. 购入 When COST is positive, it represents the spot price cost of purchasing electricity from the power grid. 购入 Negative means the spot price income from selling excess electricity to the public grid; C 电 It is the purchase price of spot electricity per unit energy.
[0081] The dispatch model of fossil energy modules, the power generation model of a single wind turbine, the power generation model of a single solar panel, the price-based demand response model and the alternative demand response model are comprehensively considered to solve the objective function, and the energy selling price is constrained by the following constraints:
[0082] Selling price constraints: The peak electricity price is higher than the plain electricity price, the plain electricity price is higher than the valley electricity price, the peak electricity price is lower than the peak electricity price upper limit, and the valley electricity price is lower than the minimum electricity price lower limit; the selling price of heating per unit power is higher than the minimum value and lower than the maximum value; the selling price of natural gas per unit power is higher than the minimum value and lower than the maximum value;
[0083] Solve the maximum profit objective function of the virtual power plant under the constraints, and record the natural gas volume consumed per unit time, the quality of coal consumed per unit time, and the peak electricity price C in the maximum profit solution. 高峰 、Plain electricity price C 平原 and valley electricity price C 低谷 , recorded as the optimal scheduling solution;
[0084] The scheduling simulation module establishes a convolutional neural network to simulate the optimal scheduling plan.
[0085] See also Figure 4 As shown in Figure 2, a four-layer convolutional neural network model is established, where f(x) and g(x) are dynamic convolution kernels, and λ A,B , B,C and λ C,D is a dynamic weight factor. X1, X2, ..., Xa are input values, and Y1, Y2 and Y3 are output values.
[0086] See also Figure 5 As shown in Figure 1, the input values X1, X2, ..., Xa of the four-layer convolutional neural network are a conditional matrix; each conditional matrix includes a benefit sub-matrix, a cost sub-matrix, an energy layout sub-matrix, and a response compliance sub-matrix. The benefit sub-matrix includes the peak electricity price C 高峰 、Plain electricity price C 平原 , Off-peak electricity price C 低谷 、The selling price of heating per unit power COST 热 、COST of natural gas per unit power 气 and the purchase price of spot electricity per unit energy C 电 ; The cost sub-matrix includes the purchase price of unit mass coal C 煤 , the purchase price of natural gas per unit volume C 天然气 , the standard operation and maintenance cost of a single wind turbine λ1, the standard operation and maintenance cost of a single solar panel λ2; the energy layout sub-matrix includes the coal mass consumed per unit time in the fossil energy module λ 煤 , the volume of natural gas consumed per unit time in the fossil energy module λ 天然气 , the number of wind turbines i and the number of solar panels j; the response obedience submatrix comes from the response obedience matrix E1 in the price-type demand response model, and the output values include the virtual power plant revenue Y1, the daily total load Y2, and the spot electricity price cost Y3 purchased from the power grid.
[0087] See also Figure 6 As shown in the figure, convolutional layer 1 and convolutional layer 2 perform data pooling after completing the convolution calculation to extract matrix features.
[0088] The training steps of the four-layer convolutional neural network are as follows:
[0089] S1: Network initialization, initialization of f(x), g(x), λ A,B , B,C and λ C,D .
[0090] S2: Collect data samples, and retrieve data from the operator subsystem and user subsystem at preset intervals to generate a condition matrix and corresponding output values. The condition matrix and the corresponding output values are divided into two categories, which are used for neural network training and comparison of prediction results respectively;
[0091] S3: Calculate the prediction error e
[0092] Where Yd is the actual output value, is the calculated value of the neural network;
[0093] S4: Correct the dynamic weight and convolution kernel, and correct the dynamic weight and convolution kernel by the prediction error e until the prediction error e is less than a preset threshold;
[0094] S5: The process ends when the prediction error e is less than the preset threshold of e, otherwise, the process returns to step S3.
[0095] After the training is completed, the conditional matrix generated by the optimal scheduling solution is substituted for calculation to obtain the result and Obtain the virtual power plant revenue Y1, daily total load Y2, and spot electricity price cost purchased from the power grid in the optimal dispatching plan. Calculate the verification feasibility index θ, where k is the preset influencing factor. When the verification feasibility index is greater than the preset threshold, the simulation verification is considered complete and the optimal scheduling plan is sent to the virtual power plant dispatch center.
[0096] Sudden adjustments to electricity prices, heating prices, and natural gas prices will have a significant impact on the consumption habits of energy consumers. Sudden increases or decreases in the production of coal and natural gas in power plants will cause a large number of equipment and personnel to be idle in a short period of time. At the same time, a large-scale expansion of wind turbine power stations and solar photovoltaic power stations in a short period of time will greatly increase management costs and construction costs. Therefore, the immediate implementation of the best optimization plan will increase the instability of the virtual power plant supply and demand system, and will affect the scheduling optimization effect. The implementation of the optimal scheduling plan should be a gradual and slow process, gradually reaching the optimal scheduling plan in a small number of times, and generating a new optimal scheduling plan during the adjustment process.
[0097] The virtual power plant dispatch center obtains the peak electricity price C in the optimal dispatch plan after simulation verification 高峰 、Plain electricity price C 平原 , Off-peak electricity price C 低谷 、The selling price of heating per unit power COST 热、COST of natural gas per unit power 气 , coal mass consumed per unit time λ 煤 , the volume of natural gas consumed per unit time λ 天然气 , the number of wind turbines i and the number of solar panels j. Generate the target data vector A = (C 高峰 , C 平原 , C 低谷 , COST 热 , COST 气 ,λ 煤 ,λ 天然气 , i, j).
[0098] Get current dispatch data, including current peak electricity prices Current Plain Electricity Price Current low electricity prices Current selling price of heating per unit power Current price of natural gas per unit power The coal mass consumed per unit time at present The volume of natural gas consumed per unit time The current number of wind turbines i0 and solar panels j0 generates the status data vector
[0099] By formula Calculate a single adjustment value where ΔX is ΔC 高峰 , ΔC 平原 , ΔC 低谷 , ΔCOST 热 , ΔCOST 气 , Δλ 煤 , Δλ 天然气 , Δi, Δj, represent single adjustment values, X is the element in the target data vector A; X 0 is the element in the current data vector; ε is the preset optimization speed factor. Generate adjustment data vector (ΔC 高峰 , ΔC 平原 , ΔC 低谷 , ΔCOST 热 , ΔCOST 气 , Δλ 煤 , Δλ 天然气 , Δi, Δj). Each optimization adjustment increases or decreases the corresponding value according to the value of the element in the adjustment data vector, and a total of ε optimization adjustments are performed.
[0100] See also Figure 7 As shown, the multi-virtual power plant joint module integrates multiple virtual power plants into the public grid to jointly play a role in scheduling optimization, export excess power to the public grid, and purchase power from the public grid when power generation is insufficient.
[0101] See also Figure 8 As shown, a virtual power plant energy optimization scheduling method includes the following steps:
[0102] Step 1: Use the energy production method of cogeneration to establish a scheduling model for the fossil energy module;
[0103] Step 2: Establish a Welbull probability distribution model of wind speed and a power generation model of a single wind turbine;
[0104] Step 3: Establish a beta probability model of light intensity and a power generation model of a single solar panel;
[0105] Step 4: Establish a price-based demand response model and an alternative demand response model;
[0106] Step 5: Obtain sensor monitoring data and substitute it into each model to solve the maximum benefit objective function and obtain the optimal scheduling plan;
[0107] Step 6: Establish a convolutional neural network to simulate the optimal scheduling solution;
[0108] Step 7: Generate an adjustment data vector according to the optimal scheduling solution;
[0109] Step 8: Perform scheduling optimization according to the adjusted data vector;
[0110] Step 9: Combine multiple virtual power plants.
[0111] It should be understood that the terms "include" and "comprising" used in the specification and claims of the present disclosure indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.
[0112] It should also be understood that the terms used in this disclosure are only for the purpose of describing specific embodiments and are not intended to limit the disclosure. As used in this disclosure and the claims, the singular forms "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise. It should also be further understood that the term "and / or" used in this disclosure and the claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes these combinations;
[0113] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.
Claims
1. A virtual power plant energy optimization scheduling system, including a fossil energy module, a renewable energy module, a demand response load module, a prediction module and a scheduling simulation module, Features: The fossil energy module adopts the energy production method of cogeneration, in which the power generation efficiency of the generator turbine converting natural gas into electricity, the heat production efficiency of the generator turbine converting natural gas into heat energy, the heat production efficiency of the thermal boiler converting coal into heat energy, the efficiency of the low-temperature waste heat power generation device converting heat energy into electricity, the efficiency of heat recovery of waste heat boilers, the standard calorific value of natural gas power generation per unit volume, and the standard value of heat production per unit mass of coal are obtained. All data are comprehensively considered to establish the scheduling model of the fossil energy module: The renewable energy module includes energy production methods for wind power generation and photovoltaic power generation, and establishes a Welbu probability distribution model for wind speed, a power generation model for a single wind turbine, a Beta probability model for light intensity, and a power generation model for a single solar panel; The specific process of establishing the Welbu probability distribution model of wind speed and the power generation model of a single wind turbine is as follows: Preset the initial scale parameter c and shape parameter k to establish the Welbu probability distribution model of wind speed At preset intervals, wind speed data is obtained through wind speed sensors arranged on each wind turbine, all wind speed data are recorded and fitted with the Welbuh probability distribution model, and the scale parameter c and shape parameter k are determined by the maximum likelihood estimation method; a power generation model of a single wind turbine is established by comprehensively considering the wind turbine number, wind speed, minimum design wind speed, maximum design wind speed, rated power generation of the wind turbine, and actual power generation of the wind turbine; The specific process of establishing the beta probability model of light intensity and the power generation model of a single solar panel is as follows: Preset the initial shape parameter α and shape parameter β to establish the beta probability model of light intensity: Obtain the maximum historical light intensity rmax of the area where the solar panel is located in a year, obtain the light intensity r through the light sensor at preset time intervals, record all light intensity data and fit it with the beta probability model of light intensity, and determine the values of shape parameters one and two through the maximum likelihood estimation method; comprehensively consider the energy conversion efficiency of the solar panel, the area of the solar panel and the light intensity, and establish a power generation model for a single solar panel; The demand response load module obtains peak power, plain power, valley power, peak electricity price adjustment, plain electricity price adjustment, valley electricity price adjustment, peak power change value, plain power change value and valley power change value through calculation, establishes a price-type demand response model and solves the response obedience matrix; then queries the unit energy standard calorific value of electricity, the unit energy standard calorific value of steam and the unit energy standard calorific value of natural gas, and establishes an alternative demand response model; The specific process of establishing a price-based demand response model and solving the response compliance matrix is as follows: Preset Response Compliance Matrix A price-based demand response model is established by comprehensively considering peak electricity price, plain electricity price, valley electricity price, peak power, plain power, valley power, peak electricity price adjustment, plain electricity price adjustment, valley electricity price adjustment, peak power change, plain power change and valley power change. An approximate solution of the response obedience matrix E is obtained based on a vector autoregression model. The process of establishing an alternative demand response model is as follows: Preset the replaced energy, the alternative energy, the load change of replaced energy A and the load change of alternative energy B; obtain the conversion coefficient from replaced energy A to alternative energy B, the utilization efficiency of replaced energy and the utilization efficiency of alternative energy, and comprehensively consider all data to establish an alternative demand response model; The constraints for establishing the alternative demand response model are: Where ΔDAmin is the minimum value of the replaced energy A; ΔDAmax is the maximum value of the replaced energy A; ΔDBmin is the minimum value of the replaced energy B; ΔDBmax is the maximum value of the replaced energy B; The prediction module sets up the maximum profit objective function of the virtual power plant, and comprehensively considers the dispatch model of the fossil energy module, the power generation model of a single wind turbine, the power generation model of a single solar panel, the price-based demand response model, and the alternative demand response model to solve the objective function; The maximum profit objective function of the virtual power plant is specifically: Among them PRO 电 Profits from selling electricity to users; PRO 热 Profits from selling heating energy to users; PRO 热 The profit from selling natural gas to users; COST 热 The selling price of heating per unit power, COST 气 COST is the price of natural gas per unit power; 煤,气 is the purchase cost of coal and natural gas, C 煤 is the purchase price of coal per unit mass, C 天然气 is the purchase price of natural gas per unit volume; COST 维护 is the maintenance cost of wind turbines and solar panels, λ1 is the standard operation and maintenance cost of a single wind turbine, and λ2 is the standard operation and maintenance cost of a single solar panel; COST 购入 is the spot price of electricity purchased from the power grid. 购入 When COST is positive, it represents the spot price cost of purchasing electricity from the power grid. 购入 When it is negative, it represents the spot electricity price income obtained by selling excess electricity to the public grid; C 电 It is the purchase price of spot electricity per unit energy; The constraints of the objective function are: the peak electricity price is greater than the plain electricity price, the plain electricity price is greater than the valley electricity price, the peak electricity price is lower than the peak electricity price upper limit, and the valley electricity price is lower than the minimum electricity price lower limit; the unit power heating price is greater than the minimum value and less than the maximum value; the unit power natural gas price is greater than the minimum value and less than the maximum value; The scheduling simulation module establishes a convolutional neural network to simulate the optimal scheduling plan.
2. A virtual power plant energy optimization scheduling system according to claim 1, It is characterized in that It also includes a load monitoring module, a virtual power plant dispatch center, and a multi-virtual power plant joint module; The load monitoring module obtains the power readings of the smart meters of each user of the virtual power plant, calculates the sum of all power readings to obtain the load power at that time, and generates a curve of load power changing with time, which is recorded as the daily load curve; a peak load and a valley load are set in the daily load curve, and the part of the load power higher than the peak load is recorded as the peak load, the part of the load power between the peak load and the valley load is recorded as the level of electricity consumption, and the part of the load power less than the valley load is recorded as the valley load; The virtual power plant dispatching center obtains the peak electricity price, plain electricity price, valley electricity price, unit power heating selling price, unit power natural gas selling price, coal mass consumed per unit time, natural gas volume consumed per unit time, number of fans and number of solar panels in the optimal dispatching scheme after simulation verification, and generates a target data vector; obtains the current dispatching data, including the current peak electricity price, the current plain electricity price, the current valley electricity price, the current unit power heating selling price, the current unit power natural gas selling price, the current coal mass consumed per unit time, the current natural gas volume consumed per unit time, the current number of fans and number of solar panels, and generates a status data vector; obtains an adjustment data vector by calculating the target data vector and the status data vector, and performs dispatching optimization according to the adjustment data vector; The multi-virtual power plant joint module integrates multiple virtual power plants into the public grid to jointly play a role in scheduling optimization, export excess power to the public grid, and purchase power from the public grid when power generation is insufficient.
3. A virtual power plant energy optimization scheduling system according to claim 1, It is characterized in that The energy production method of combined heat and power is as follows: The electric energy generated by natural gas in the power turbine is output as electricity; the steam generated by coal in the thermal boiler is output as heat; the turbine waste heat generated by the power turbine is divided into turbine waste heat 1 and turbine waste heat 2, turbine waste heat 1 is output to the low-temperature waste heat power generation device, and turbine waste heat 2 is output to the waste heat boiler; the low-temperature waste heat power generation device realizes heat-to-electricity conversion through organic Rankine cycle technology, converts turbine waste heat 1 into electric energy, and incorporates it into the power output line of the power turbine; the waste heat boiler realizes waste heat reuse through heat energy recovery, converts turbine waste heat 2 into steam, and incorporates it into the thermal output line of the thermal boiler.
4. A virtual power plant energy optimization scheduling system according to claim 1, It is characterized in that The specific process of the convolutional neural network simulating the optimal scheduling solution is as follows: Establish a four-layer convolutional neural network model, set dynamic convolution kernel, dynamic weight factor, input value and output value; The input values of the preset four-layer convolutional neural network are several conditional matrices; each conditional matrix includes a benefit sub-matrix, a cost sub-matrix, an energy layout sub-matrix and a response compliance sub-matrix; the benefit sub-matrix includes peak electricity price, plain electricity price, valley electricity price, unit power heating selling price, unit power natural gas selling price and unit energy spot electricity purchase price; the cost sub-matrix includes the purchase price of unit mass coal, the purchase price of unit volume natural gas, the standard operation and maintenance cost of a single wind turbine, and the standard operation and maintenance cost of a single solar panel; the energy layout sub-matrix includes the coal mass consumed per unit time in the fossil energy module, the natural gas volume consumed per unit time in the fossil energy module, the number of wind turbines i and the number of solar panels j; the response compliance sub-matrix comes from the response compliance matrix E1 in the price-based demand response model, and the output values include the virtual power plant revenue, the total daily load, and the spot electricity price cost purchased from the power grid; The training steps of the four-layer convolutional neural network are as follows: S1: Network initialization, initialization of winter convolution kernel and dynamic weight factor; S2: Collect data samples, and retrieve data from the operator subsystem and the user subsystem at preset intervals to generate a condition matrix and corresponding output values; The condition matrix and the corresponding output values are divided into two categories, which are used for neural network training and comparison of prediction results respectively; S3: Calculate the prediction error through operation; S4: Correcting the dynamic weights and convolution kernels, and correcting the dynamic weights and convolution kernels by prediction errors until the prediction error is less than a preset threshold; S5: The process ends when the prediction error is less than the preset threshold, otherwise, it returns to step S3; After the training is completed, the condition matrix generated by the optimal scheduling plan is substituted for calculation to obtain the output result; the virtual power plant revenue, daily total load and spot electricity price cost purchased from the power grid in the optimal scheduling plan are obtained; the verification feasibility index is obtained through formula calculation; when the verification feasibility index is greater than the preset threshold, the simulation verification is judged to be completed, and the optimal scheduling plan is sent to the virtual power plant dispatching center.
5. A virtual power plant energy optimization scheduling system according to claim 1, It is characterized in that The method of using the system includes the following steps: Step 1: Use the energy production method of cogeneration to establish a scheduling model for the fossil energy module; Step 2: Establish a Welbull probability distribution model of wind speed and a power generation model of a single wind turbine; Step 3: Establish a beta probability model of light intensity and a power generation model of a single solar panel; Step 4: Establish a price-based demand response model and an alternative demand response model; Step 5: Obtain sensor monitoring data and substitute it into each model to solve the maximum benefit objective function and obtain the optimal scheduling plan; Step 6: Establish a convolutional neural network to simulate the optimal scheduling solution; Step 7: Generate an adjustment data vector according to the optimal scheduling solution; Step 8: Perform scheduling optimization according to the adjusted data vector; Step 9: Combine multiple virtual power plants.
