Virtual power plant resource aggregation method based on multi-time scale characteristic modeling

Abstract

The invention discloses a virtual power plant resource aggregation method based on multi-time scale characteristic modeling, and the method comprises the steps: taking historical data as a sample for training to obtain a preliminary long and short term memory neural network prediction model, and the historical data comprises original load data, meteorological data, date factors and peak and valley electricity price factors; inputting the weather information of the to-be-predicted time and the time type information of whether the to-be-predicted time is a workday into the prediction model to obtain a prediction result; and analyzing a prediction result, and if the accuracy requirement is not met, performing supplementary training and updating on the long-short-term memory neural network prediction model through a new training sample. According to the virtual power plant resource aggregation method, the accuracy of evaluating the adjustable capability of the virtual power plant aggregation resources is improved, the method is more reliable when participating in power grid business, and the assessment risk is reduced.

Description

Technical Field

[0001] This invention belongs to the field of energy management technology, and particularly relates to a virtual power plant resource aggregation method based on multi-timescale characteristic modeling. Background Technology

[0002] Virtual power plants utilize advanced communication technologies to control distributed resources such as distributed power sources, energy storage systems, and controllable loads, participating in grid operation and market transactions. Existing technologies typically use simple parameters (such as maximum / minimum output and adjustability) to characterize distributed resources, or simply classify and aggregate them. However, the output of distributed resources (especially photovoltaic, wind power, and electric vehicles) exhibits strong randomness and volatility, with its response characteristics dynamically changing with factors such as weather, user behavior, and equipment status. This makes existing static aggregation models unable to accurately describe the real-time controllable potential of resource clusters. Summary of the Invention

[0003] This application proposes a virtual power plant resource aggregation method based on multi-timescale characteristic modeling to address the aforementioned technical problems. By establishing a model through data, confidence intervals of predicted power and load values ​​are obtained. Minkowski theorem is used to aggregate the adjustable intervals of distributed resources, and the adjustable intervals of aggregated resources are further aggregated to obtain the regulation potential. Finally, the embedded superbox method is used to approximate the solution to obtain the reliable quantitative interval of the regulation potential of the virtual power plant.

[0004] The specific technical solution is as follows: A virtual power plant resource aggregation method based on multi-timescale characteristic modeling, wherein the method is as follows: A preliminary long short-term memory neural network prediction model is obtained by training based on historical data as samples. The historical data includes raw load data, meteorological data, date factors, and peak-valley electricity price factors. Input the meteorological information of the time to be predicted and the time type information, such as whether it is a weekday, into the prediction model to obtain the prediction result; The prediction results are analyzed, and if the accuracy requirements are not met, the long short-term memory neural network prediction model is supplemented and updated using new training samples.

[0005] Furthermore, the method for analyzing the prediction results is as follows: The distribution functions of the expected predicted value and the prediction error are obtained through the prediction model; By using the inverse distribution function method, the chance constraint is transformed into an equivalent deterministic constraint, and the confidence interval of the source-load prediction value is obtained. The aggregated adjustable power space of a virtual power plant is calculated based on confidence level.

[0006] Furthermore, the method of transforming chance constraints into equivalent deterministic constraints using the inverse distribution function method is as follows: use chance constraints with confidence levels to represent the predicted values ​​of power generation and consumption, and then use the inverse distribution function method to transform the chance constraints into deterministic constraints. At a specific confidence level α, the range of variation of power generation and consumption as random variables is expressed by the following chance constraint: ; In the above formula, α is the confidence level; Pr(·) represents the probability of the random variable; and j represents the type of distributed power resources.

[0007] Furthermore, the upper and lower bounds of the generated and consumed power at confidence level α are constrained as follows: .

[0008] Furthermore, the method for calculating the aggregated adjustable power space of the virtual power plant based on confidence level is as follows: the Minkowski method is used to aggregate the adjustable power space of the resources, and then the embedded superbox method is used for approximate solution to obtain the aggregated adjustable power range of the virtual power plant.

[0009] The beneficial effects of this invention are as follows: The virtual power plant resource aggregation method of the present invention improves the accuracy of assessing the adjustability of aggregated virtual power plant resources, making it more reliable when participating in power grid business and reducing assessment risks. Attached Figure Description

[0010] Figure 1 The diagram shown illustrates the step change in the charging and discharging of the air conditioning load. Figure 2 This is a schematic diagram of the basic unit of a neural network; Figure 3 The figure shows the baseline confidence interval for the load. Figure 4 The diagram shows Minkowski and a schematic diagram; Figure 5 The diagram shown is a schematic of the embedded superbox method; Figure 6 The maximum adjustable power is shown. Figure 7 The maximum adjustable power is shown. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments.

[0012] In this application, the characteristics of distributed resources are first analyzed and modeled.

[0013] Air conditioners have thermal equilibrium inertia, and can schedule start-stop cycles while meeting user comfort, thus participating in the day-ahead dispatch of the power grid as a demand-side controllable resource. Large-scale electric vehicles, as a distributed energy storage unit, can use V2G technology to achieve friendly interaction with the smart grid and participate in grid dispatch by changing charging and discharging modes. Distributed energy storage systems can achieve power regulation by utilizing the spatiotemporal regulation characteristics of energy storage and release.

[0014] To simplify the constraints of these distributed energy resources and streamline the power aggregation of the integrated electric-hydrogen virtual power plant, a reduced-order virtual energy storage (VS) model is introduced.

[0015] ; in, and These represent the virtual storage (SOVC) states at times t and t-1, respectively. It is the charging and discharging power; and express The lower and upper bounds; and This reflects the upper and lower bounds of SOVC; η is the charge / discharge efficiency.

[0016] The VS model for electric vehicles. Assuming the electric vehicle starts charging when connected to the grid, the upper limit of the virtual state of charge (SOVC) at time t can be calculated; under the condition of meeting the minimum charging amount when the vehicle is disconnected from the grid, the lower limit of the virtual state of charge at time t can be calculated. Through analysis, the VS model for electric vehicles is established as follows: in, It is the SOVC of electric vehicles; , These are the upper and lower limits of SOVC for electric vehicles; It is the equivalent charging and discharging power of an electric vehicle; η EV The charging and discharging efficiency coefficient for electric vehicles; , This refers to the maximum and minimum charging / discharging power of electric vehicles.

[0017] The VS parameter of the air conditioner, taking into account the heat transfer between indoors and outdoors, adopts the classic equivalent thermal parameter (ETP) modeling method to simulate the heat transfer process. Through analysis and modeling, the operating power of the air conditioner exhibits a step change between its maximum and minimum values, such as... Figure 1 As shown.

[0018] Based on the analysis, the equivalent VS output constraint for electric vehicles is established as follows: R, C, λ, and the synthesis parameters are obtained through a bottom-up approach.

[0019] Considering the uncertainties at both the power supply and load ends, renewable energy output and electricity consumption are both random variables, which can be described as the sum of the expected predicted value and the prediction error. Based on a long short-term memory neural network, we use historical data to drive the generation of the distribution functions of the expected predicted value and the prediction error. Then, using the inverse distribution function method, we transform the chance constraint into an equivalent deterministic constraint, obtaining the confidence interval of the source-load predicted value.

[0020] A virtual power plant resource aggregation method based on multi-timescale characteristic modeling, wherein the method is as follows: A preliminary long short-term memory neural network prediction model was obtained by training based on historical data as samples. This historical data includes raw load data, meteorological data, date factors, and peak / valley electricity price factors. The long short-term memory neural network prediction model is as follows: Figure 2 As shown; Input the meteorological information of the time to be predicted and the time type information, such as whether it is a weekday, into the prediction model to obtain the prediction result; The prediction results are analyzed, and if the accuracy requirements are not met, the long short-term memory neural network prediction model is supplemented and updated using new training samples.

[0021] In this application, considering the volatility of renewable energy power generation and the uncertainty of load users' energy consumption behavior, opportunity constraints with confidence levels are used to represent the predicted values ​​of power generation and consumption, and then the inverse distribution function method is used to transform the opportunity constraints into deterministic constraints.

[0022] Furthermore, the method for analyzing the prediction results is as follows: The distribution functions of the expected predicted value and the prediction error are obtained through the prediction model; By using the inverse distribution function method, the chance constraint is transformed into an equivalent deterministic constraint, and the confidence interval of the source-load prediction value is obtained. The aggregated adjustable power space of a virtual power plant is calculated based on confidence level.

[0023] Furthermore, the method of transforming chance constraints into equivalent deterministic constraints using the inverse distribution function method is as follows: use chance constraints with confidence levels to represent the predicted values ​​of power generation and consumption, and then use the inverse distribution function method to transform the chance constraints into deterministic constraints. At a specific confidence level α, the range of variation of power generation and consumption as random variables is expressed by the following chance constraint: ; In the above formula, α is the confidence level; Pr(·) represents the probability of the random variable; and j represents the type of distributed power resources.

[0024] According to the rules for calculating random variables, the optimistic value of the random variable ξ is... and pessimistic value The cumulative distribution function (CDF) is defined as follows: Under the constraints of the above formula, It is clearly a random variable with multiple possibilities. and However, in the VS model, only the pessimistic or optimistic values ​​are of interest. Therefore, assuming the CDF of the random variable is symmetric, minimizing the pessimistic value yields: Based on the definition of pessimism, the upper and lower bounds of a random variable at confidence level α can be calculated: Where Φ -1 (y) represents the inverse function of y=Φ(x).

[0025] After comparing the actual and predicted values, the prediction error approximately follows a normal distribution. Based on the definition of the arcnormal distribution function, the upper and lower limits of the random variable are determined as follows: .

[0026] Figure 3 The forecast range of the load baseline at different confidence levels is shown. Under the influence of uncertainty, the load baseline changes from a single curve to an error band, and the range of the error band expands as the confidence level increases.

[0027] Based on the obtained confidence intervals for distributed resources, it is known that the adjustable power of distributed resources exhibits time coupling and uncertainty. Therefore, at a certain confidence level, the adjustable power can be visualized as a convex polyhedron with power at each time period as the coordinate axis. Since the adjustable power of distributed resources is additivity, the Minkowski sum (symbolized as "...") is introduced. The aggregate adjustable power space of the virtual power plant is calculated using the formula П1, which represents the power feasible region П1 for time periods t and t+1. Taking П2 as an example, the operational rules of the Minkowski sum are shown as follows: Figure 4 .

[0028] As the time interval and distributed resources increase, the dimension of solving the aggregated power distribution increases dramatically, causing the corresponding Minkowski sum to become an NP-hard problem that cannot be solved directly. Therefore, the embedded superbox method is used to approximate the solution for the aggregated power interval, such as... Figure 5 As shown.

[0029] By using Minkowski algorithm and aggregating the adjustable space of resources, and then approximating the solution using the embedded superbox method, the aggregated adjustable power range of the virtual power plant was obtained. Figure 6 , Figure 7 The maximum adjustable power and maximum adjustable power of the virtual power plant are shown respectively at a 50% confidence level.

[0030] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it.

Claims

1. A virtual power plant resource aggregation method based on multi-timescale characteristic modeling, characterized in that, The method is as follows: A preliminary long short-term memory neural network prediction model is obtained by training based on historical data as samples. The historical data includes raw load data, meteorological data, date factors, and peak-valley electricity price factors. Input the meteorological information of the time to be predicted and the time type information, such as whether it is a weekday, into the prediction model to obtain the prediction result; The prediction results are analyzed, and if the accuracy requirements are not met, the long short-term memory neural network prediction model is supplemented and updated using new training samples.

2. The virtual power plant resource aggregation method based on multi-timescale characteristic modeling according to claim 1, characterized in that, The method for analyzing the prediction results is as follows: The distribution functions of the expected predicted value and the prediction error are obtained through the prediction model; By using the inverse distribution function method, the chance constraint is transformed into an equivalent deterministic constraint, and the confidence interval of the source-load prediction value is obtained. The aggregated adjustable power space of a virtual power plant is calculated based on confidence level.

3. The virtual power plant resource aggregation method based on multi-timescale characteristic modeling as described in 2, characterized in that, The method of transforming chance constraints into equivalent deterministic constraints using the inverse distribution function method is as follows: use chance constraints with confidence levels to represent the predicted values ​​of power generation and consumption, and then use the inverse distribution function method to transform the chance constraints into deterministic constraints. At a specific confidence level α, the range of variation of power generation and consumption as random variables is expressed by the following chance constraint: ; In the above formula, α is the confidence level; Pr(·) represents the probability of the random variable; and j represents the type of distributed power resources.

4. The virtual power plant resource aggregation method based on multi-timescale characteristic modeling as described in 3, characterized in that, The upper and lower bounds of the generated and consumed power at confidence level α are constrained as follows: 。 5. The virtual power plant resource aggregation method based on multi-timescale characteristic modeling as described in 4, characterized in that, The method for calculating the aggregated adjustable power space of a virtual power plant based on confidence is as follows: the Minkowski method is used to aggregate the adjustable power space of resources, and then the embedded superbox method is used for approximate solution to obtain the aggregated adjustable power range of the virtual power plant.

Images