New energy-stored energy joint operation strategy optimization method based on uncertainty modeling and distributed robust optimization and application thereof

Abstract

The invention discloses a new energy-energy storage combined operation strategy optimization method based on uncertainty modeling and distributed robust optimization. The method comprises the following steps: a) carrying out data acquisition and preprocessing on power generation data and power market prices of a new energy plant station; b) using a fuzzy set to express uncertainty factors in the data; c) based on the probability distribution of the uncertainty data, establishing an optimization model of new energy-energy storage joint operation through a distribution robust optimization method; d) modeling and solving the optimization problem by using a deep neural network or a convolutional neural network; and e) adjusting a scheduling strategy of the energy storage facility according to an optimization result, ensuring maximum consumption of new energy in an uncertain environment, reducing wind and light abandoning phenomena, and improving robustness of power grid scheduling. The method has the advantages that the intelligent level of scheduling is remarkably improved, and the risk and cost of a traditional scheduling method are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of power system joint operation optimization, and in particular to a joint operation optimization method for a new energy power generation and energy storage system and its application. Background Art

[0002] Wind power generation has become an important part of the global energy transformation due to its renewable and low-emission characteristics. However, wind power output has significant uncertainty and volatility, and traditional deterministic forecasting methods are difficult to fully reflect the uncertainty of future output. Especially in the electricity market, power generators need to submit power generation plans in advance, and failure to accurately forecast may lead to economic losses. As the proportion of new energy (such as wind power, solar energy, etc.) in the power system gradually increases, the volatility and intermittency of energy such as wind power and photovoltaics have brought great challenges to the dispatch and management of the power grid. Traditional point forecasting methods usually ignore the probabilistic characteristics of new energy power generation, resulting in the inability to effectively deal with uncertainty and volatility.

[0003] Existing methods mainly rely on point prediction and deterministic optimization models. These methods are difficult to accurately handle the volatility and randomness of renewable energy output, especially when faced with factors such as price fluctuations in the electricity spot market and grid load uncertainty. They cannot effectively improve the robustness of grid scheduling. Therefore, how to accurately model the probability distribution of renewable energy power generation and optimize scheduling using distributed robust optimization methods is the key to solving this problem. The present invention proposes a new energy-energy storage joint operation strategy optimization method based on uncertainty modeling and distributed robust optimization. Summary of the invention

[0004] The main purpose of the present invention is to provide a new energy-energy storage joint operation strategy optimization method based on uncertainty modeling and distributionally robust optimization and its application in order to solve the above technical problems. This method optimizes the joint operation strategy of new energy plants and energy storage systems through uncertainty modeling, distributionally robust optimization (DRO) and neural network solving methods to improve the absorption capacity of new energy, stabilize the load of the power grid and reduce the phenomenon of wind and solar power abandonment.

[0005] The technical solution adopted by the present invention to solve the above problems is as follows:

[0006] The optimization method of new energy-energy storage joint operation strategy based on uncertainty modeling and distributed robust optimization is as follows:

[0007] a) Collect and pre-process the power generation data and electricity market prices of new energy plants, including cleaning, denoising, standardization and normalization of multi-dimensional data of wind speed, light, temperature, humidity and electricity market prices;

[0008] b) Use fuzzy sets to represent the uncertainty factors in the data, and use generative adversarial networks (GANs) or variational autoencoders (VAEs) to model the probability distribution of PV output and electricity prices, and generate probability distribution samples of uncertain data;

[0009] c) Based on the probability distribution of uncertainty data, an optimization model for the joint operation of new energy and energy storage is established through the distributed robust optimization (DRO) method. The optimization model contains an objective function and uncertainty constraints. The objective function is used to maximize benefits or minimize costs. The constraints include wind power, photovoltaic power generation output, charging and discharging capacity of energy storage facilities, and grid load.

[0010] d) Using a deep neural network (DNN) or a convolutional neural network (CNN) to model and solve the above optimization problem, the input of the network is uncertainty data, the output is the scheduling strategy of the energy storage equipment, and the network weights are optimized through the back propagation algorithm to obtain the optimal scheduling strategy for energy storage facilities;

[0011] e) Adjust the dispatch strategy of energy storage facilities according to the optimization results to ensure the maximum absorption of new energy under uncertain environment, reduce the phenomenon of wind and solar power abandonment, and improve the robustness of grid dispatch.

[0012] Preferably, the objective function of the distributed robust optimization method is:

[0013]

[0014] Among them, R(u) represents the profit function, u is the decision variable (the scheduling strategy of the energy storage equipment), λ is the risk control coefficient, and Risk(u) is the risk function, which is used to quantify the uncertainty of the scheduling strategy.

[0015] Preferably, the uncertainty constraint of the distributed robust optimization method is:

[0016]

[0017] Among them, Y wind For wind power output, is the possible output range of wind power, and ∈ is the tolerance.

[0018] Preferably, the objective function of the generative adversarial network (GAN) is:

[0019]

[0020] Among them, G is the generator, D is the discriminator, and p data (x) is the distribution of real data, p z (z) is the distribution of latent variables and G(z) is the generator output.

[0021] Preferably, the loss function of the variational autoencoder (VAE) is:

[0022]

[0023] Among them, D KL is the Kullback-Leibler divergence, p θ (z) is the probability distribution of the generative model, q φ (z|x) is the posterior distribution of the encoder.

[0024] Preferably, the objective function of the neural network training is:

[0025]

[0026] Among them, θ is the parameter of the neural network, R objective To optimize the objective function, RiskFactor(u(θ)) is a risk assessment function, which aims to optimize the scheduling strategy of energy storage facilities through neural networks.

[0027] Preferably, the energy storage facility scheduling strategy includes charging and discharging decisions of energy storage equipment, energy storage power management, and grid load scheduling decisions, which can dynamically adjust the operation mode of energy storage facilities under changing power demand and new energy power generation conditions, maximize new energy consumption, and ensure stable operation of the power grid.

[0028] An application of new energy-energy storage joint operation strategy optimization based on uncertainty modeling and distributed robust optimization, including:

[0029] a) Uncertainty modeling module, which is used to model the uncertainty of wind power, photovoltaic power generation data and electricity market prices, and uses the generative adversarial network (GAN) and variational autoencoder (VAE) methods to generate the probability distribution of uncertainty data;

[0030] b) The distributed blue-rod optimization module is used to establish an optimization model for the joint operation of new energy and energy storage, and optimize decisions through the distributed blue-rod optimization method to maximize system benefits and reduce wind and solar power abandonment;

[0031] c) Energy storage scheduling module, which uses deep neural network (DNN) or convolutional neural network (CNN) to intelligently optimize the scheduling strategy of energy storage equipment and dynamically adjust it according to the grid load and renewable energy power generation conditions.

[0032] Preferably, it also includes a real-time data acquisition module for collecting real-time power generation data, electricity market price data and grid load data of new energy plants and stations, and inputting them into the uncertainty modeling module in real time for updating and adjustment.

[0033] The beneficial effects that can be achieved by the present invention using the above technical solution are:

[0034] The present invention adopts advanced machine learning technologies such as generative adversarial networks (GAN) and variational autoencoders (VAE), which can accurately model the probability distribution of renewable energy generation and electricity market prices, so as to better capture and describe uncertainty factors. Based on the distributionally robust optimization method, it can ensure that the energy storage scheduling strategy is always in the optimal state in the face of uncertainty, and effectively balance the volatility of renewable energy generation. By solving the distributionally robust optimization problem through a neural network, the scheduling strategy of the energy storage equipment can be dynamically adjusted, the intelligence level of scheduling can be significantly improved, and the risk and cost of traditional scheduling methods can be reduced. The present invention provides a new energy-energy storage joint operation strategy optimization method based on uncertainty modeling and distributionally robust optimization, combined with technologies such as generative adversarial networks (GAN), variational autoencoders (VAE) and deep neural networks (DNN), which can effectively improve the absorption capacity of renewable energy generation, reduce the phenomenon of power abandonment, and optimize the scheduling strategy of the energy storage system, and has broad application prospects. DETAILED DESCRIPTION

[0035] Uncertainty Modeling

[0036] Both renewable energy generation and electricity market prices are affected by a variety of uncertain factors, such as weather changes, wind speed fluctuations, market price fluctuations, etc. The present invention first represents these uncertain factors as fuzzy sets, and learns the probability distribution of uncertain factors through machine learning methods such as generative adversarial networks (GANs) and variational autoencoders (VAEs).

[0037] The uncertainty factor means:

[0038] Multidimensional data such as wind speed, sunlight, temperature, and electricity price are represented as fuzzy sets. Assume that X = {x1, x2, …, x n} are uncertain inputs (such as wind speed, light, temperature, etc.), which can be expressed as fuzzy sets in is the membership function of the fuzzy set at time t, is the range of each uncertainty factor. The uncertainty is expressed by the membership function of the fuzzy set:

[0039]

[0040] in, is the membership function of the fuzzy set, x i (t) is the i-th uncertainty factor at time t, α i is the fuzzy parameter.

[0041] Using GAN to learn wind power output Y wind and electricity price Pprice For example, for the prediction of photovoltaic output, suppose the distribution obtained through GAN training is P(Y pv |X), which represents the probability distribution of photovoltaic power generation under given input feature X. The objective function of GAN is:

[0042]

[0043] Among them, G is the generator, D is the discriminator, and p data (x) is the distribution of real data, p z (z) is the distribution of latent variables and G(z) is the generator output.

[0044] To further improve the uncertainty sample, we use VAE to probabilistically model PV power output and electricity prices and learn the distribution of its latent variable z:

[0045]

[0046] Among them, D KL is the Kullback-Leibler divergence, p θ (z) is the probability distribution of the generative model, q φ (z|x) is the posterior distribution of the encoder.

[0047] To deal with uncertainty, this paper uses the distributed robust optimization (DRO) method to optimize the operation of the station. This method solves an optimization problem with uncertainty constraints to ensure that the combined operation of new energy and energy storage can achieve the optimal goal under different market and power generation conditions. Assuming that the goal of the new energy plant is to maximize revenue or reduce power curtailment, its objective function is:

[0048]

[0049] Among them, R(u) represents the profit function, u is the decision variable (the scheduling strategy of the energy storage equipment), λ is the risk control coefficient, and Risk(u) is the risk function, which is used to quantify the uncertainty of the scheduling strategy.

[0050] The output of wind power and photovoltaic power generation is random, and the constraints can be expressed as:

[0051]

[0052] Among them, Y wind For wind power output, is the possible output range of wind power, and ∈ is the tolerance.

[0053] The charging and discharging constraints on energy storage devices can be expressed as:

[0054] Emin ≤E t ≤E max

[0055] Among them, E t represents the power of the energy storage device at time t, E min and E max are the minimum and maximum power of the energy storage facility respectively.

[0056] The optimization objective of the distribution is:

[0057]

[0058] Among them, f(u,z) is the objective function of the system (such as cost, benefit, etc.), z is the uncertainty factor (such as wind speed, light, electricity price, etc.), is the probability distribution set of uncertain data.

[0059] In order to effectively solve the distributionally robust optimization problem, especially when dealing with nonlinear and high-dimensional data, the present invention adopts a deep neural network (DNN) to express and solve the distributionally robust optimization problem.

[0060] Use deep neural networks (DNN) or convolutional neural networks (CNN) to represent the objective function and constraints of distributed robust optimization. The network inputs are uncertain factors such as wind speed, light, and electricity prices, and the output is the optimal scheduling strategy for energy storage devices. The training goal of the network is to minimize the loss function L(θ) so that the scheduling strategy u(θ) of the energy storage device can meet the constraints of distributed robust optimization during the optimization process:

[0061]

[0062] The neural network is trained using the back-propagation algorithm, and the network weights θ are adjusted so that the output energy storage facility calling strategy u(θ) can achieve optimal performance under different market environments and uncertainty conditions.

Claims

1. A new energy-energy storage joint operation strategy optimization method based on uncertainty modeling and distributed robust optimization, characterized in that: The steps are as follows: a) Collect and pre-process the power generation data and electricity market prices of new energy plants, including cleaning, denoising, standardization and normalization of multi-dimensional data of wind speed, light, temperature, humidity and electricity market prices; b) Use fuzzy sets to represent the uncertainty factors in the data, and use generative adversarial networks or variational autoencoders to model the probability distribution of photovoltaic output and electricity prices to generate probability distribution samples of uncertain data; c) Based on the probability distribution of uncertainty data, an optimization model for the joint operation of new energy and energy storage is established through the distributed robust optimization method. The optimization model contains an objective function and uncertainty constraints. The objective function is used to maximize benefits or minimize costs. The constraints include wind power, photovoltaic power generation output, charging and discharging capacity of energy storage facilities, and grid load. d) Using a deep neural network or a convolutional neural network to model and solve the above optimization problem, the input of the network is uncertainty data, the output is the scheduling strategy of the energy storage equipment, and the network weight is optimized by the back propagation algorithm to obtain the optimal scheduling strategy of the energy storage facility; e) Adjust the dispatch strategy of energy storage facilities according to the optimization results to ensure the maximum absorption of new energy under uncertain environment, reduce the phenomenon of wind and solar power abandonment, and improve the robustness of grid dispatch.

2. The new energy-energy storage joint operation strategy optimization method based on uncertainty modeling and distributed robust optimization according to claim 1 is characterized in that: The objective function of the distributed robust optimization method is: Among them, R(u) represents the profit function, u is the decision variable, λ is the risk control coefficient, and Risk(u) is the risk function, which is used to quantify the uncertainty of the scheduling strategy.

3. The new energy-energy storage joint operation strategy optimization method based on uncertainty modeling and distributed robust optimization according to claim 2 is characterized in that: The uncertainty constraints of the distributed robust optimization method are: Among them, Y wind For wind power output, is the possible output range of wind power, and ∈ is the tolerance.

4. The new energy-energy storage joint operation strategy optimization method based on uncertainty modeling and distributed robust optimization according to claim 1 is characterized in that: The objective function of the generative adversarial network is: Among them, G is the generator, D is the discriminator, and p data (x) is the distribution of real data, p z (z) is the distribution of latent variables and G(z) is the generator output.

5. The new energy-energy storage joint operation strategy optimization method based on uncertainty modeling and distributed robust optimization according to claim 1 is characterized in that: The loss function of the variational autoencoder is: Among them, D KL is the Kullback-Leibler divergence, p θ (z) is the probability distribution of the generative model, q φ (z|x) is the posterior distribution of the encoder.

6. The new energy-energy storage joint operation strategy optimization method based on uncertainty modeling and distributed robust optimization according to claim 1 is characterized in that: The objective function of the neural network training is: Among them, θ is the parameter of the neural network, R objective To optimize the objective function, RiskFactor(u(θ)) is a risk assessment function, which aims to optimize the scheduling strategy of energy storage facilities through neural networks.

7. The new energy-energy storage joint operation strategy optimization method based on uncertainty modeling and distributed robust optimization according to claim 1 is characterized in that: The energy storage facility scheduling strategy includes charging and discharging decisions of energy storage equipment, energy storage power management, and grid load scheduling decisions. It can dynamically adjust the operation mode of energy storage facilities under changing power demand and new energy power generation conditions, maximize new energy consumption, and ensure stable operation of the power grid.

8. A new energy-energy storage joint operation strategy optimization application based on uncertainty modeling and distributed robust optimization according to claim 1, 2, 3, 4, 5, 6 or 7, characterized in that: include: a) Uncertainty modeling module, which is used to model the uncertainty of wind power, photovoltaic power generation data and electricity market prices, and uses generative adversarial networks and variational autoencoder methods to generate the probability distribution of uncertainty data; b) The distributed blue-rod optimization module is used to establish an optimization model for the joint operation of new energy and energy storage, and optimize decisions through the distributed blue-rod optimization method to maximize system benefits and reduce wind and solar power abandonment; c) Energy storage scheduling module, which uses deep neural networks or convolutional neural networks to intelligently optimize the scheduling strategy of energy storage equipment, and dynamically adjust it according to the grid load and renewable energy power generation conditions.

9. The application of new energy-energy storage joint operation strategy optimization based on uncertainty modeling and distributed robust optimization according to claim 8 is characterized in that: It also includes a real-time data acquisition module, which is used to collect real-time power generation data, electricity market price data and grid load data of new energy plants and stations, and input them into the uncertainty modeling module in real time for updating and adjustment.

Images