An energy base scheduling method and terminal based on a distribution robust optimization method

Abstract

The application discloses an energy base scheduling method and a terminal based on a distribution robust optimization method. A deep convolutional generative adversarial network based on wasserstein distance is trained by using historical load-wind power data to generate wind power output data conforming to a real data distribution, and then a typical scenario is generated based on the wind power output data. An energy base optimization model is converted into a data-driven distribution robust optimization model based on wasserstein distance, and the data-driven distribution robust optimization model based on wasserstein distance is solved based on the typical scenario to obtain a scheduling strategy. The distribution robust optimization method can consider the situation under various extreme scenarios and make corresponding adjustment scheduling schemes. Based on the real data, the obtained scheduling strategy is not excessively influenced by the extreme scenarios so as to be excessively conservative, so that the system has better ability to cope with various situations, the robustness of the system is improved, and thus the economy and safety of the energy base operation are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to an energy base scheduling method and terminal based on a distributed robust optimization method. Background Art

[0002] Since the Industrial Revolution, the massive consumption of fossil energy has led to air pollution and climate change, posing a serious threat to the sustainable development of human society. To reduce carbon emissions in the energy sector and reduce dependence on fossil fuels, my country has actively promoted the development of renewable energy and explored new forms of energy systems. However, in some regions, there is a temporal and spatial mismatch between renewable energy output and user load demand. Simply increasing the development of renewable resources will result in a certain degree of resource waste and fail to meet growing load demands. Therefore, energy storage technology is needed to achieve spatial and temporal energy transfer, giving rise to the concept of energy bases.

[0003] Currently, commonly used energy storage methods at home and abroad include battery storage, pumped storage, and hydrogen storage. Electrochemical energy storage systems, with their flexible configuration and easy maintenance, are widely used in power systems. They can quickly respond to frequency modulation signals, reduce load peaks and valleys, and promote the consumption of renewable energy. Pumped storage uses water as an energy storage medium, converting electrical energy into potential energy to achieve the storage and management of electrical energy. It is the most technologically mature and economically efficient energy storage method. Hydrogen energy is of great significance as a secondary energy source. Through electrohydrogen production technology, surplus renewable energy can be converted into hydrogen for storage, achieving large-scale seasonal energy storage, promoting the consumption of renewable energy and the efficient use of social resources. These energy storage methods play an important role in the power system, promoting the low-carbon transformation and green development of the power energy system.

[0004] However, there are still a series of technical problems that need to be solved in the coordinated development and application of renewable energy and diversified energy storage in energy bases. These are mainly the following two problems:

[0005] Question 1: The theory of multi-energy complementary energy base is incomplete. Most existing studies break down the modeling of coupled systems into single electricity, water, and hydrogen energy systems. This analysis method cannot reflect the flexibility and economy of the coupled system.

[0006] Question 2: When it comes to the uncertainty of renewable energy output, most existing studies simply cluster existing historical data, and then input wind and solar output and load as deterministic data into the optimization model for optimization. This method cannot take into account situations that did not appear in the historical data and has poor robustness. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide an energy base scheduling method and terminal based on a distributed robust optimization method, which can ensure the economy and safety of the energy base operation.

[0008] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0009] An energy base scheduling method based on a distributed robust optimization method comprises the following steps:

[0010] Using historical load-wind power data to train a deep convolutional generative adversarial network based on Wasserstein distance to obtain a trained deep convolutional generative adversarial network based on Wasserstein distance, and using the trained deep convolutional generative adversarial network based on Wasserstein distance to generate wind power output data that obeys the real data distribution;

[0011] generating a typical scenario based on the wind power output data;

[0012] Constructing an objective function based on maximizing the net income of the energy base, establishing constraints of the energy base, and establishing an energy base optimization model based on the objective function and the constraints;

[0013] The energy base optimization model is converted into a data-driven distributed robust optimization model based on the Wasserstein distance, and the data-driven distributed robust optimization model based on the Wasserstein distance is solved based on the typical scenario to obtain a scheduling strategy.

[0014] In order to solve the above technical problems, another technical solution adopted by the present invention is:

[0015] An energy base dispatching terminal based on a distributed robust optimization method includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following steps are implemented:

[0016] Using historical load-wind power data to train a deep convolutional generative adversarial network based on Wasserstein distance to obtain a trained deep convolutional generative adversarial network based on Wasserstein distance, and using the trained deep convolutional generative adversarial network based on Wasserstein distance to generate wind power output data that obeys the real data distribution;

[0017] generating a typical scenario based on the wind power output data;

[0018] Constructing an objective function based on maximizing the net income of the energy base, establishing constraints of the energy base, and establishing an energy base optimization model based on the objective function and the constraints;

[0019] The energy base optimization model is converted into a data-driven distributed robust optimization model based on the Wasserstein distance, and the data-driven distributed robust optimization model based on the Wasserstein distance is solved based on the typical scenario to obtain a scheduling strategy.

[0020] The beneficial effects of the present invention are: a deep convolutional generative adversarial network based on the Wasserstein distance is trained using historical load-wind power data to generate wind power output data that obeys the real data distribution, and then a typical scenario is generated based on the wind power output data. The energy base optimization model constructs an objective function to maximize the net profit of the energy base, which is converted into a data-driven distributionally robust optimization model based on the Wasserstein distance, and the data-driven distributionally robust optimization model based on the Wasserstein distance is solved based on the typical scenario to obtain a scheduling strategy. The distributionally robust optimization method can take into account various uncertainties in the energy system, take into account the situations in various extreme scenarios and make corresponding adjustments to the scheduling plan, and based on historical real data, the obtained scheduling strategy will not be overly affected by extreme scenarios so that the strategy is overly conservative, so that the system is more capable of coping with various situations, and the robustness of the system is improved, thereby ensuring the economy and safety of the energy base operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a flowchart of the steps of an energy base scheduling method based on a distributed robust optimization method according to an embodiment of the present invention;

[0022] Figure 2 This is a structural diagram of an energy base dispatching terminal based on a distributed robust optimization method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] To illustrate the technical content, achieved objectives and effects of the present invention in detail, the following description is given in conjunction with the embodiments and accompanying drawings.

[0024] Please refer to Figure 1 , an energy base scheduling method based on a distributed robust optimization method, comprising the steps of:

[0025] Using historical load-wind power data to train a deep convolutional generative adversarial network based on Wasserstein distance to obtain a trained deep convolutional generative adversarial network based on Wasserstein distance, and using the trained deep convolutional generative adversarial network based on Wasserstein distance to generate wind power output data that obeys the real data distribution;

[0026] generating a typical scenario based on the wind power output data;

[0027] Constructing an objective function based on maximizing the net income of the energy base, establishing constraints of the energy base, and establishing an energy base optimization model based on the objective function and the constraints;

[0028] The energy base optimization model is converted into a data-driven distributed robust optimization model based on the Wasserstein distance, and the data-driven distributed robust optimization model based on the Wasserstein distance is solved based on the typical scenario to obtain a scheduling strategy.

[0029] From the above description, it can be seen that the beneficial effect of the present invention is that: the deep convolutional generative adversarial network based on the wasserstein distance is trained using historical load-wind power data to generate wind power output data that obeys the real data distribution, and then a typical scenario is generated based on the wind power output data. The energy base optimization model constructs an objective function to maximize the net profit of the energy base, and converts it into a data-driven distributionally robust optimization model based on the wasserstein distance. The data-driven distributionally robust optimization model based on the wasserstein distance is solved based on the typical scenario to obtain a scheduling strategy. The distributionally robust optimization method can take into account various uncertainties in the energy system, take into account the situations in various extreme scenarios and make corresponding adjustments to the scheduling plan. Moreover, based on historical real data, the obtained scheduling strategy will not be overly affected by extreme scenarios so that the strategy is overly conservative, making the system more capable of coping with various situations and improving the robustness of the system, thereby ensuring the economy and safety of the energy base operation.

[0030] Furthermore, generating a typical scenario based on the wind power output data includes:

[0031] Clustering the wind power output data using a k-means clustering algorithm to obtain cluster centers;

[0032] The cluster center is taken as a typical scenario.

[0033] From the above description, we can see that using the K-means clustering algorithm to cluster wind power output data reduces the dimension and complexity of the data, improves the model's generalization ability for data, and can extract the features of typical scenarios. These features are then passed as input to the subsequent data-driven distribution robust optimization model based on the Wasserstein distance, thereby improving the robustness of the model.

[0034] Furthermore, the method of using the trained deep convolutional generative adversarial network based on the wasserstein distance to generate wind power output data that obeys the real data distribution includes:

[0035] Noise data that obeys a random distribution is input into the trained deep convolutional generative adversarial network based on the wasserstein distance, and wind power output data that obeys the real data distribution is output.

[0036] From the above description, it can be seen that by inputting noise data that obeys a random distribution into the trained deep convolutional generative adversarial network based on the Wasserstein distance, the output wind power output data obeys the real data distribution. This fully takes into account situations that have not appeared in historical data, making the system more capable of coping with various situations and improving the robustness of the system.

[0037] Furthermore, constructing an objective function based on maximizing the net income of the energy base includes:

[0038]

[0039] Where W represents the net income of the energy base, represents the benefits of reducing carbon emissions, F Eb represents the total revenue, C Inv represents the total investment cost, C Run Indicates the equipment operation and maintenance cost, C Dep represents the total depreciation cost of equipment, C pumped represents the total cost of the pumped storage power station, represents the penalty cost for wind curtailment.

[0040] From the above description, it can be seen that the objective function is constructed by maximizing the net income of the energy base, which comprehensively considers the three aspects of the national strategic level, the operating costs of the energy base, and the operating income of the energy base. It can comprehensively measure the comprehensive benefits of the energy base, avoid the one-sidedness of a single indicator, and maximize the operating income of the base while considering the operating costs. It helps to balance the economic benefits and sustainable development of the energy base, making it economically feasible.

[0041] Furthermore, the constraints for establishing the energy base include:

[0042] Establish energy balance constraints, stable operation constraints, battery operation constraints, hydrogen energy storage system operation constraints, and pumped storage system constraints for the energy base;

[0043] Constraint conditions are generated according to the energy balance constraint, the stable operation constraint, the battery operation constraint, the hydrogen energy storage system operation constraint, and the pumped storage system constraint.

[0044] From the above description, it can be seen that establishing the energy balance constraints, stable operation constraints, battery operation constraints, hydrogen energy storage system operation constraints, and pumped storage system constraints of the energy base ensures that the technical characteristics and performance of the selected equipment meet the requirements of the energy base operation, thereby ensuring that the selected equipment can efficiently convert energy and provide stable output. Moreover, since energy bases usually require long-term operation and have high reliability requirements for energy supply, equipment-related constraints can set equipment reliability indicators to ensure that the selected equipment has sufficient service life, anti-interference ability, and fault tolerance, thereby reducing operation risks and maintenance costs.

[0045] Furthermore, converting the energy base optimization model into a data-driven distributed robust optimization model based on Wasserstein distance includes:

[0046] Simplifying the energy base optimization model to obtain a simplified energy base optimization model;

[0047] According to the strong duality theorem, the inner and outer layers of the simplified energy base optimization model are interchanged, and a data-driven distributed robust optimization model based on the Wasserstein distance is reconstructed.

[0048] From the above description, it can be seen that by simplifying the energy base optimization model, and then exchanging the inner and outer layers of the simplified energy base optimization model according to the strong duality theorem, and reconstructing the data-driven distributional robust optimization model based on the wasserstein distance, the infinite-dimensional original problem can be transformed into a finite-dimensional convex optimization problem for easy solution.

[0049] Furthermore, the energy base optimization model is simplified to obtain a simplified energy base optimization model including:

[0050]

[0051]

[0052]

[0053] Where x represents the controllable decision variables in the energy base, P 0represents the empirical distribution of the data obtained after scene generation and scene reduction, P represents the joint probability distribution, B() represents the wasserstein ball, c() represents the objective function, P wind represents wind power, Q() represents an abstract inequality, H() represents an abstract equality, α represents the first dual variable, β represents the second dual variable, and ω represents the third dual variable.

[0054] From the above description, it can be seen that the energy base optimization model is simplified to reduce the solution complexity and improve the solution speed.

[0055] Furthermore, the simplified energy base optimization model is interchanged between inner and outer layers according to the strong duality theorem, and a data-driven distributed robust optimization model based on the Wasserstein distance is reconstructed, including:

[0056]

[0057] Where ξ represents a random variable, Represents a set of random variables, and u() represents a function of the third dual variable.

[0058] From the above description, it can be seen that according to the strong duality theorem, the inner and outer layers of the simplified energy base optimization model are interchanged and reconstructed, and finally a finite-dimensional convex optimization problem is obtained, so that it can be better solved and the optimal energy base scheduling strategy can be obtained.

[0059] Furthermore, the data-driven distributed robust optimization model based on the wasserstein distance is solved based on the typical scenario to obtain a scheduling strategy including:

[0060] Based on the typical scenario, the gurobi solver is used to solve the data-driven distributed robust optimization model based on the wasserstein distance to obtain the scheduling plan and the output of each component of the energy base.

[0061] From the above description, we can see that using the Gurobi solver to solve the data-driven distributed robust optimization model based on the Wasserstein distance based on typical scenarios is more efficient and reliable.

[0062] Please refer to Figure 2 Another embodiment of the present invention provides an energy base scheduling terminal based on a distributionally robust optimization method, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, each step of the energy base scheduling method based on the distributionally robust optimization method is implemented.

[0063] The energy base scheduling method and terminal based on the distributed robust optimization method of the present invention can be applied to energy base optimization scheduling scenarios, which are described below through specific implementation methods:

[0064] Please refer to Figure 1 , embodiment 1 of the present invention is:

[0065] An energy base scheduling method based on a distributed robust optimization method comprises the following steps:

[0066] S1. Use historical load-wind power data to train a deep convolutional generative adversarial network based on the Wasserstein distance to obtain a trained deep convolutional generative adversarial network based on the Wasserstein distance, and use the trained deep convolutional generative adversarial network based on the Wasserstein distance to generate wind power output data that obeys the real data distribution.

[0067] Specifically, historical load-wind power data is used to train a deep convolutional generative adversarial network based on the Wasserstein distance. When the "Nash equilibrium" is reached, the training is completed, and a trained deep convolutional generative adversarial network based on the Wasserstein distance is obtained. Noise data that obeys a random distribution is input into the trained deep convolutional generative adversarial network based on the Wasserstein distance, and wind power output data that obeys the real data distribution is output.

[0068] The deep convolutional generative adversarial network is a deep learning model that can implicitly represent the probability distribution of a dataset by training on historical datasets. The generator and discriminator need to continuously learn and optimize themselves throughout the game process and eventually reach a "Nash equilibrium" state. When the generative neural network (GAN) is used to generate distributed power generation scenarios, a set of probability distributions p z The random noise data z of (z) is used as the input of the generator, and the real data obeys the probability distribution p real (x). The generated data sample output by the generator is G(z), and its probability distribution is p G (z). The goal of the generator is to make the probability distribution of the generated data samples close to the historical data through training. The output of the generator is the generated data sample, whose probability distribution is p G (z). The role of the generator is to make the probability distribution of the generated data samples as close as possible to the historical data through training. For the discriminator, the input data can be historical data or generated data, and the discriminator should distinguish them as accurately as possible. Generator L G and the discriminator L D The loss functions are as follows:

[0069]

[0070]

[0071] The traditional deep convolutional generative adversarial network adopts JS divergence (Jensen-Shannon divergence) and KL divergence (Kullback-Leibler divergence), but the second one has a fatal flaw. When the two distributions do not overlap at all, the JS divergence and KL divergence will be 0, which causes the loss function to become a constant and the gradient disappearance problem occurs. The advantage of Wasserstein distance is that even if the two distributions do not overlap, it can still accurately reflect the distance between them. Therefore, the present invention uses Wasserstein distance to improve the traditional deep convolutional generative adversarial network, that is, the deep convolutional generative adversarial network based on Wasserstein distance. The Wasserstein distance is expressed as follows:

[0072]

[0073] The objective function of the deep convolutional generative adversarial network based on the wasserstein distance is:

[0074]

[0075] Where, represents the gradient of x', and K represents the setting parameter.

[0076] S2. Generate a typical scenario based on the wind power output data, specifically including S21-S22:

[0077] S21. Cluster the wind power output data using a k-means clustering algorithm to obtain cluster centers.

[0078] S22: The cluster center is used as a typical scenario, that is, a typical wind power output scenario.

[0079] S3. Constructing an objective function based on maximizing the net income of the energy base, establishing constraints for the energy base, and establishing an energy base optimization model based on the objective function and the constraints, specifically including S31-S34:

[0080] S31. Construct an objective function based on maximizing the net income of the energy base, specifically:

[0081]

[0082] Where W represents the net income of the energy base, represents the benefits of reducing carbon emissions, FEb represents the total revenue, C Inv represents the total investment cost, C Run Indicates the equipment operation and maintenance cost, C Dep represents the total depreciation cost of equipment, C pumped represents the total cost of the pumped storage power station, represents the penalty cost for wind curtailment.

[0083] Global climate change has become a major concern for humanity. Establishing carbon emission targets and implementing carbon reduction policies can help reduce greenhouse gas emissions and mitigate their impact on global climate change. These policies can encourage businesses and individuals to adopt energy-saving and emission-reduction measures, improve resource efficiency, reduce environmental pollution and damage, and achieve sustainable development. Furthermore, carbon reduction policies can encourage business transformation and upgrading, promote economic restructuring, foster green development and a low-carbon economy, and enhance industrial competitiveness and sustainable development. Therefore, establishing carbon emission targets and implementing carbon reduction policies are important initiatives that contribute to environmental protection, economic development, and the fulfillment of international responsibilities.

[0084] The benefits of reducing carbon emissions are:

[0085]

[0086] In the formula, θ represents the average carbon emissions per kilowatt-hour, represents the carbon emission price, P e Indicates the amount of electricity generated by the energy base.

[0087] The total revenue is:

[0088]

[0089]

[0090]

[0091] Where N represents the number of typical weeks, and the total revenue of N weeks is used to approximate the total revenue of the whole year. e1,i It indicates the revenue gained by the energy base from selling excess electricity to the grid, R e2,i represents the emission reduction benefits generated by wind power in the energy base, P net,i,t represents the electric power sold in the energy base at time t, K e1,i,t represents the on-grid electricity price at time t, μ loss Represents the transmission loss coefficient, K c Indicates the price of reducing carbon emissions at the energy base, in RMB / kg, K f Indicates the conversion ratio of carbon emissions to coal burning in thermal power generation, in kg / kWh.

[0092] The total investment cost is:

[0093]

[0094]

[0095] Where C Inv,k represents the total investment cost of the equipment, where k = 1 represents electrolyzer, k = 2 represents fuel cell, k = 3 represents hydrogen storage tank, k = 4 represents pumped storage, k = 5 represents battery, r represents the discount rate, m represents the operating life of the energy base, P ele,max represents the investment capacity of the electrolytic cell, P fc,max represents the fuel cell investment capacity, P tan,max Indicates the investment capacity of the hydrogen storage tank, P wat,max represents the investment capacity of pumped storage, P bat,max represents the investment capacity of the battery, e1 represents the unit investment cost of the electrolyzer, e2 represents the unit investment cost of the fuel cell, e3 represents the unit investment cost of the hydrogen storage tank, e4 represents the unit investment cost of pumped storage, and e5 represents the unit investment cost of the battery.

[0096] The equipment operation and maintenance costs are:

[0097]

[0098] Where Q k It indicates the percentage of the average annual operation and maintenance cost of each equipment to the initial investment cost of the equipment.

[0099] The total depreciation cost of the equipment is:

[0100]

[0101] Where C R Indicates the residual value, which is 10%C Inv .

[0102] The total cost of the pumped storage power station is:

[0103]

[0104] Where C pump_cap Indicates the cost per unit volume of the reservoir, in yuan / m3, V pump_cap Indicates the maximum capacity of the downstream reservoir to be constructed, C power Indicates the maintenance cost of the pumped turbine per installed unit power, in RMB / kW, P power Indicates the power of the pump turbine of the pumped storage system, T a Indicates the total number of years the energy base has been in operation, Com_pc Indicates the maintenance cost per unit volume of the reservoir, in yuan / m3, V pump_pp Indicates the maximum volume of the upstream reservoir, C om_pp V represents the investment cost of the pump turbine per unit capacity of the pumped storage system. power represents the total investment in pumped turbine capacity, C rep_pp It represents the cost of replacing the turbine of the pumped storage power station in the future due to failure, in RMB / kW, T pump It indicates the life of the pumped storage turbine equipment in the station, in years.

[0105] The penalty cost for wind curtailment is:

[0106]

[0107] Where η represents the wind curtailment penalty coefficient, P wind,i,t Indicates the wind power curtailment.

[0108] S32. Establish energy balance constraints, stable operation constraints, battery operation constraints, hydrogen energy storage system operation constraints, and pumped storage system constraints for the energy base.

[0109] The energy balance constraint is:

[0110]

[0111]

[0112] Where, It represents the sum of wind, fire, water and core active power, represents the total active power of chemical batteries, hydrogen energy storage and pumped storage, P grid Indicates the active power of the grid, Indicates the total active power of the load, represents the sum of wind, fire, water and core reactive power, represents the sum of reactive power of chemical batteries, hydrogen energy storage and pumped storage, Q grid Indicates the reactive power of the grid, Indicates the total reactive power of the load.

[0113] The stable operation constraints are:

[0114] U min ≤U t ≤U max ;

[0115] β min ≤β t ≤β max ;

[0116] fmin ≤f t ≤f max ;

[0117] Where U min Indicates the minimum voltage for the power grid to operate safely and stably, U t Indicates the voltage at which the power grid can operate safely and stably, U max Indicates the maximum voltage at which the power grid can operate safely and stably, β min Indicates the minimum phase angle for safe operation of the power grid, β t Indicates the phase angle at which the power grid can operate safely, β max Indicates the maximum phase angle at which the power grid can operate safely, f min Indicates the minimum grid frequency at which the grid can operate safely, f t Indicates the grid frequency at which the grid can operate safely, f max Indicates the maximum grid frequency at which the grid can operate safely.

[0118] In order to protect the cycle life of the battery, the energy storage charge state must be strictly controlled to avoid overcharging and over-discharging. During use, the guidance of the battery management system should be followed to strictly control the upper and lower limits of the charge state. The energy storage device needs to ensure sufficient output power, and the maximum output energy must also meet the load requirements. In addition, to avoid sudden large power loss when the load is using electricity, such as starting a large motor, which causes a sharp drop in power quality, the composite energy storage device must be able to quickly release large power to support the system, while ensuring that the battery storage energy does not exceed the respective maximum power limit. Therefore, the battery operation constraints are:

[0119] S OC,min ≤S OC,i ≤S OC,max ;

[0120]

[0121] P os,i +P bat,i ≥ΔP max ;

[0122] P bat,min ≤P bat,i ≤P bat,max ;

[0123] Where S OC,min Indicates the lower limit of energy storage charge state, S OC,i Indicates the energy storage charge state, S OC,max Indicates the upper limit of the energy storage state of charge. In an optional embodiment, S OC,min 20% to 30%, S OC,max 80% to 100%, S OC,i+1Indicates the energy storage charge state at the next moment, ΔE bat,i Indicates the energy change of battery storage, n indicates the climbing coefficient, E k Indicates the rated capacity of the battery, P bat,i represents the rated power of the battery, T represents the scheduling time period, P os,i Represents the total power that can be generated by other energy storages, P bat,i Indicates the power that the battery can generate, ΔP max Indicates the maximum instantaneous power loss, P bat,min Indicates the lower limit of the power that the battery can generate, P bat,max Indicates the upper limit of the power that the battery can produce.

[0124] As one of the core devices in the hydrogen storage system, the hydrogen storage tank is used to store the hydrogen produced by the electrolyzer when there is energy surplus. The hydrogen tank can achieve energy translation in time and space. To extend the service life of the equipment and ensure stable hydrogen production and power generation, the start and stop time and power variation of the electrolyzer and fuel cell are constrained. The operating constraints of the hydrogen energy storage system are as follows:

[0125]

[0126]

[0127] P ele,i,t =η ele V he,i,t ;

[0128]

[0129]

[0130]

[0131]

[0132]

[0133] Where, P h_max Indicates the maximum outflow or inflow rate of the hydrogen storage tank, P h,i,t It indicates the amount of hydrogen flowing out of or into the hydrogen storage tank at time t. When it is a positive value, it means that hydrogen is released from the hydrogen storage tank. When it is a negative value, it means that hydrogen is filled into the hydrogen storage tank. h,i,t Indicates the amount of gas in the hydrogen storage tank at time t, E h,i,t-1 Indicates the amount of hydrogen contained in the hydrogen storage tank at time t-1, P h,t Indicates the amount of hydrogen produced per unit time by the electrolyzer, E h_max Indicates the maximum hydrogen storage capacity of the hydrogen storage tank. Indicates the rated minimum operating power of the electrolytic cell, Pele,i,t Indicates the operating power of the electrolyzer, Indicates the rated maximum operating power of the electrolytic cell, η ele Indicates the efficiency of P2G per unit volume of hydrogen, V he,i,t represents the volume of hydrogen produced by electrolytic cell i at time t, P fc,i,t represents the power of the fuel cell, η fc represents the fuel cell efficiency, Indicates the maximum rated power of the fuel cell, V hf,i,t Indicates the fuel cell output, U ele,i,t represents the start and stop variables of the i-th electrolytic cell at time t, U ele,i,t-1 represents the start and stop variables of the i-th electrolytic cell at time t-1, U ele,i,p Indicates the intermediate variable of start-stop control, p indicates the intermediate variable, SU ele Indicates the minimum start-up time of the electrolyzer, SD ele Indicates the minimum stop time of the electrolytic cell, U fc,i,t-1 represents the start-stop variable of the i-th fuel cell at time t-1, U fc,i,t represents the start-stop variable of the i-th fuel cell at time t, U fc,i,p Indicates the start-stop control intermediate variable, SU fc Indicates the minimum start-up time of the fuel cell, SD fc Indicates the minimum stop time of the fuel cell, P ele,i,t-1 Represents the operating power of the electrolytic cell at time t-1, RU ELE Indicates the rising power limit of the electrolyzer, RD ELE Indicates the reduced power limit of the electrolyzer, P fc,i,t-1 Represents the power of the fuel cell at time t-1, RU FC Indicates the rising power limit of the fuel cell, RD FC Indicates the power reduction limit of the fuel cell. In this embodiment, SU ele , SD ele SU fc , SD fc 1 hour, RU ELE , RD ELE , RU FC , RD FC Each is 20% of the maximum power of the corresponding device.

[0134] The pumped storage system constraints are mainly composed of three parts: power constraint, reservoir water capacity constraint and water inflow and outflow constraint per unit time.

[0135] The power constraint is:

[0136]

[0137] Where, γ H represents the 0-1 water discharge state control variable of the pumped storage power station, P psmax represents the installed capacity of the turbine unit, P p,i,t represents the power generated by the turbine unit when the water is released, γ P represents the 0-1 pumping state control variable of the pumped storage power station. Under normal circumstances, pumping and releasing cannot be carried out at the same time, so this constraint is used to constrain the control variable. H and γ P Cannot take 1 at the same time.

[0138] The reservoir water capacity constraint is:

[0139]

[0140] Where c psmin It represents the ratio of the minimum storage capacity of the upstream reservoir of the pumped storage system to the maximum energy, E psmax represents the maximum storage energy of the upstream reservoir of the pumped storage system, E ps,i,t represents the water storage capacity of the upstream reservoir of the pumped storage system at time t, E ps,i,t-1 represents the water storage capacity of the upstream reservoir of the pumped storage system at time t-1, α P Indicates the energy conversion efficiency when releasing water, α H It represents the energy conversion efficiency during pumping, and ΔT represents the unit interval time of optimized operation.

[0141] The water inlet and outlet constraints per unit time are:

[0142]

[0143] Where N t Represents the total number of discrete points for the optimization run.

[0144] S33. Generate constraint conditions according to the energy balance constraint, the stable operation constraint, the battery operation constraint, the hydrogen energy storage system operation constraint, and the pumped storage system constraint.

[0145] S34. Establish an energy base optimization model based on the objective function and the constraint conditions.

[0146] S4. Converting the energy base optimization model into a data-driven distributed robust optimization model based on Wasserstein distance, and solving the data-driven distributed robust optimization model based on Wasserstein distance based on the typical scenario to obtain a scheduling strategy, specifically including S41-S43:

[0147] S41. Simplify the energy base optimization model to obtain a simplified energy base optimization model, specifically:

[0148]

[0149]

[0150]

[0151] Where x represents the controllable decision variables in the energy base, P 0 represents the empirical distribution of the data obtained after scene generation and scene reduction, P represents the joint probability distribution, B() represents the wasserstein ball, c() represents the objective function, P wind represents wind power, Q() represents an abstract inequality, H() represents an abstract equality, α represents the first dual variable, β represents the second dual variable, and ω represents the third dual variable.

[0152] S42. According to the strong duality theorem, the inner and outer layers of the simplified energy base optimization model are interchanged, and a data-driven distributed robust optimization model based on the Wasserstein distance is reconstructed, specifically:

[0153]

[0154] Where ξ represents a random variable, Represents a set of random variables, and u() represents a function of the third dual variable.

[0155] The data-driven distributionally robust optimization model based on Wasserstein distance adopts the first-order Wasserstein distance as a measurement method for measuring the empirical distribution and the true distribution in distributionally robust optimization, specifically:

[0156]

[0157] Where π represents the joint probability distribution of ξ1 and ξ2, and its marginal distributions are P and P 0 as follows:

[0158]

[0159]

[0160] Where, P wind,i Indicates the true value of wind power, P 0 wind,i Indicates historical data value of wind power.

[0161] The Wasserstein distance is used to define a sphere with the empirical distribution of historical data as the center ε and the Wasserstein distance as the radius, which constrains the wind power output around the fuzzy set of the empirical distribution:

[0162] B(P 0 )={W(P,P 0 )≤ε};

[0163] But since W(P,P 0 ) It is an infinite-dimensional problem in itself and cannot be solved directly as a constraint. It needs to be equivalently transformed into a finite-dimensional convex optimization problem before it can be solved. Therefore, the S41-S42 operations are performed.

[0164] S43. Based on the typical scenario, the gurobi solver is used to solve the data-driven distributed robust optimization model based on the wasserstein distance to obtain the scheduling plan and the output of each component of the energy base.

[0165] Please refer to Figure 2 , the second embodiment of the present invention is:

[0166] An energy base scheduling terminal based on a distributionally robust optimization method includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, each step of the energy base scheduling method based on the distributionally robust optimization method in Example 1 is implemented.

[0167] In summary, the present invention provides an energy base scheduling method and terminal based on the distributional robust optimization method, which uses historical load-wind power data to train a deep convolutional generative adversarial network based on the wasserstein distance to generate wind power output data that obeys the real data distribution, and then generates typical scenarios based on the wind power output data. The energy base optimization model constructs an objective function to maximize the net profit of the energy base, and converts it into a data-driven distributional robust optimization model based on the wasserstein distance. The data-driven distributional robust optimization model based on the wasserstein distance is solved based on typical scenarios to obtain a scheduling strategy. The distributional robust optimization method can take into account various uncertainties in the energy system, take into account the situations in various extreme scenarios and make corresponding adjustments to the scheduling plan, and based on historical real data, the obtained scheduling The strategy will not be overly affected by extreme scenarios and become overly conservative, making the system more capable of coping with various situations and improving the robustness of the system, thereby ensuring the economy and safety of the energy base operation; in addition, the noise data that obeys the random distribution is input into the trained deep convolutional generative adversarial network based on the wasserstein distance, and the wind power output data that obeys the real data distribution is output, which fully takes into account the situations that have not appeared in the historical data, making the system more capable of coping with various situations and improving the robustness of the system; and the energy base optimization model is simplified, and then the inner and outer layers of the simplified energy base optimization model are interchanged according to the strong duality theorem, and the data-driven distributional robust optimization model based on the wasserstein distance is reconstructed, which can transform the infinite-dimensional original problem into a finite-dimensional convex optimization problem for easy solution.

[0168] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent transformations made using the contents of the present invention's description and drawings, or directly or indirectly applied in related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. An energy base scheduling method based on a distributed robust optimization method, characterized in that: Including steps: Using historical load-wind power data to train a deep convolutional generative adversarial network based on Wasserstein distance to obtain a trained deep convolutional generative adversarial network based on Wasserstein distance, and using the trained deep convolutional generative adversarial network based on Wasserstein distance to generate wind power output data that obeys the real data distribution; generating a typical scenario based on the wind power output data; Constructing an objective function based on maximizing the net income of the energy base, establishing constraints of the energy base, and establishing an energy base optimization model based on the objective function and the constraints; Converting the energy base optimization model into a data-driven distributed robust optimization model based on Wasserstein distance, and solving the data-driven distributed robust optimization model based on Wasserstein distance based on the typical scenario to obtain a scheduling strategy; The objective function constructed by maximizing the net benefit of the energy base includes: ； Where W represents the net income of the energy base, Represents the benefits of reducing carbon emissions, represents the total revenue, represents the total investment cost, Indicates the equipment operation and maintenance cost, represents the total depreciation cost of the equipment, represents the total cost of the pumped storage power station, represents the penalty cost for wind curtailment; The constraints for establishing the energy base include: Establish energy balance constraints, stable operation constraints, battery operation constraints, hydrogen energy storage system operation constraints, and pumped storage system constraints for the energy base; generating a constraint condition according to the energy balance constraint, the stable operation constraint, the battery operation constraint, the hydrogen energy storage system operation constraint, and the pumped storage system constraint; The converting of the energy base optimization model into a data-driven distributed robust optimization model based on the Wasserstein distance comprises: Simplifying the energy base optimization model to obtain a simplified energy base optimization model; According to the strong duality theorem, the simplified energy base optimization model is exchanged between the inner and outer layers, and a data-driven distributed robust optimization model based on the Wasserstein distance is reconstructed; The energy base optimization model is simplified to obtain a simplified energy base optimization model including: Where x represents the controllable decision variables in the energy base, P 0 represents the empirical distribution of the data obtained after scene generation and scene reduction, P represents the joint probability distribution, B() represents the wasserstein ball, c() represents the objective function, P wind represents wind power, Q() represents an abstract inequality, H() represents an abstract equality, α represents the first dual variable, β represents the second dual variable, and ω represents the third dual variable.

2. The energy base scheduling method based on the distributed robust optimization method according to claim 1 is characterized in that: The typical scenario generated based on the wind power output data includes: Clustering the wind power output data using a k-means clustering algorithm to obtain cluster centers; The cluster center is taken as a typical scenario.

3. The energy base scheduling method based on the distributed robust optimization method according to claim 1 is characterized in that: The method of using the trained deep convolutional generative adversarial network based on the wasserstein distance to generate wind power output data that obeys the real data distribution includes: Noise data that obeys a random distribution is input into the trained deep convolutional generative adversarial network based on the wasserstein distance, and wind power output data that obeys the real data distribution is output.

4. The energy base scheduling method based on the distributed robust optimization method according to claim 1 is characterized in that: The simplified energy base optimization model is interchanged with the inner and outer layers according to the strong duality theorem, and the data-driven distributed robust optimization model based on the Wasserstein distance is reconstructed, including: ； Where, represents a random variable, Denotes a set of random variables, and u() denotes a function of the third dual variable.

5. The energy base scheduling method based on the distributed robust optimization method according to claim 1 is characterized in that: The data-driven distributed robust optimization model based on the Wasserstein distance is solved based on the typical scenario to obtain a scheduling strategy including: Based on the typical scenario, the gurobi solver is used to solve the data-driven distributed robust optimization model based on the wasserstein distance to obtain the scheduling plan and the output of each component of the energy base.

6. An energy base dispatching terminal based on a distributed robust optimization method, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the processor implements each step of the energy base scheduling method based on the distributional robust optimization method according to any one of claims 1 to 5.

Images