A Dynamic Peak Shaving Method and System for Virtual Power Plants Based on Neural Networks

Abstract

The present invention discloses a dynamic peak shaving method and system for a virtual power plant based on a neural network, which relates to the technical field of power grids and includes: obtaining system dispatching instructions on the power supply side and the demand side of the virtual power plant in real time, constructing data of controllable power generation resources at different time scales for the system dispatching instructions, and establishing an adjustable power quantity model of power generation resources at different time scales; analyzing the adjustable power quantity characteristics of the power supply side and the demand side through the adjustable power quantity model of power generation resources, establishing a virtual power plant resource aggregation model, and obtaining multi-level optimized dispatching data of the virtual power plant; using a dynamic peak shaving strategy to construct an objective function for the multi-level optimized dispatching data, introducing dynamic constraint conditions to construct an output dispatching model for the multi-level optimized dispatching data, and obtaining real-time power dynamic peak shaving data, which can make the load curve smoother, improve the peak shaving effect, and obtain the adjustable flexibility of the virtual power plant by integrating flexible adjustable resources, ensuring the stability of power grid operation.

Description

Technical Field

[0001] The present invention relates to the technical field of virtual power plant dynamic peak shaving, and specifically relates to a method and system for virtual power plant dynamic peak shaving based on a neural network. Background Technique

[0002] As a power source coordination management system, a virtual power plant can combine various clean energy sources, controllable loads, and energy storage systems installed dispersedly in the distribution network as a special power plant to participate in power grid operation and electricity market transactions, thereby effectively reducing the difficulty of power grid dispatching, improving the consumption of renewable energy, and ensuring the safe and stable operation of the power system. However, in the actual operation process of the virtual power plant at present, it mainly focuses on economic benefits and the consumption of renewable energy, and fails to fully exert its potential for low-carbon emission reduction. Carbon emissions need to consider carbon costs, and the carbon reduction benefits of renewable energy are quantified, so as to effectively stimulate the enthusiasm and potential of each carbon emission unit for emission reduction, which will inevitably have an important impact on the arrangement of the virtual power plant operation and dispatching plan. The existing methods for virtual power plant dynamic peak shaving still have the following problems:

[0003] (1) With the continuous increase in the grid connection ratios of low-schedulability resources such as wind power and photovoltaic power, as well as nuclear power units, the peak-valley difference of the load increases accordingly, and the power grid peak shaving situation becomes increasingly severe. The demand for nuclear power units to participate in system peak shaving operation is becoming more urgent, resulting in a large gap in the grid's low valley peak shaving capacity. Relying solely on the annual load tracking and weekly regulation operation of nuclear power units is not sufficient to meet the peak shaving demand of the system;

[0004] (2) Due to the randomness of the output of uncertain power sources such as wind power and photovoltaic power, as well as the existence of random variables such as load prediction errors, the dispatching operation of virtual power plant dynamic peak shaving is uncontrollable. How to properly handle uncertain variables has become the key to ensuring the reliability of power grid operation. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for virtual power plant dynamic peak shaving based on a neural network to solve the technical problems in the prior art that there is a large gap in the grid's low valley peak shaving capacity, relying solely on the annual load tracking and weekly regulation operation of nuclear power units is not sufficient to meet the peak shaving demand of the system, and the dispatching operation of virtual power plant dynamic peak shaving is uncontrollable caused by uncertain variables.

[0006] To solve the above technical problems, the present invention specifically provides the following technical solutions:

[0007] In the first aspect of the present invention, a method for virtual power plant dynamic peak shaving based on a neural network is provided, including the following steps:

[0008] Obtain the system scheduling instructions of the power supply side and the demand side of the virtual power plant in real time, construct the controllable power generation resource data at different time scales for the system scheduling instructions, analyze the power characteristics of the controllable power generation resource data, and establish the adjustable power quantity model of the power generation resources at different time scales;

[0009] Analyze the adjustable power quantity characteristics of the power supply side and the demand side through the adjustable power quantity model of the power generation resources, converge the characteristics of different power resources, establish the resource aggregation model of the virtual power plant, and obtain the multi-level optimal scheduling data of the virtual power plant;

[0010] Adopt a dynamic peak shaving strategy to construct an objective function for the multi-level optimal scheduling data, introduce dynamic constraint conditions to construct an output scheduling model for the multi-level optimal scheduling data, and obtain real-time power dynamic peak shaving data.

[0011] As a preferred solution of the present invention, obtaining the system scheduling instructions of the power supply side and the demand side of the virtual power plant in real time, and constructing the controllable power generation resource data at different time scales for the system scheduling instructions, including:

[0012] Statistical characteristic indexes of the power generation resources corresponding to the system scheduling instructions, and classifying the characteristic indexes into the regulation rate of the power generation resources, the adjustable power, and the adjustable power quantity;

[0013] Establish a power generation resource adjustable power quantity diagram with the power axis as the vertical axis and the time axis as the horizontal axis according to the characteristic indexes at different time scales;

[0014] The power generation resource adjustable power quantity diagram starts from the initial power generation operation point (0, p 0 ), and at any time t = τ 1 , t = τ 2 respectively calculate the upward adjustment power quantity E 1 , the downward adjustment power quantity E 2 , and their expressions are:

[0015]

[0016] Among them, p 1 , p 2 respectively represent the power quantity power values at the moments of τ 1 , τ 2 , respectively represent the maximum adjustable power and the minimum adjustable power during the adjustment process of the power quantity, p max represents the upper limit value of power adjustment, p min represents the lower limit value of power adjustment, P c represents the power generation resource regulation efficiency value;

[0017] According to the power generation resource adjustable power quantity diagram, from τ 1 to τ 2Within the time scale, regulate the power generation according to the maximum power, and obtain the data of controllable power generation resources at different time scales.

[0018] As a preferred embodiment of the present invention, analyze the power characteristics of the controllable power generation resource data, and establish an adjustable power generation resource model at different time scales, including:

[0019] Divide the controllable power generation resource data into M types γ = {γ 1 , γ 2 , …, γ i , …, γ M}, i ∈ (1, M), and use the single power generation data of the γ i type as the data center γ DC ;

[0020] Statistically analyze the data centers of M types γ, and set the regulation probability for each data center to obtain the regulation probability distribution corresponding to the controllable power generation resource data. Its expression is:

[0021]

[0022] Among them, D represents the amplitude of power fluctuation generated during power regulation in the power market;

[0023] Maintain the total power balance on the power supply side and the demand side according to the regulation probability distribution, and respond to the interruptible load on the demand side through the load characteristics during the power time shift process. Its expression is:

[0024] E 1 (Δτ, P L ) = Δτ × min((P es - P e ), (E e - E emax ));

[0025] Among them, Δτ represents the time difference between τ 1 , τ 2 moments, P es represents the maximum charging power included in the controllable power generation resource data, P e represents the current power generation power, E e represents the current energy storage power, and E emax represents the maximum adjustable power within the time period of τ 1 , τ 2 ;

[0026] Analyze the dispatchable power resources on the demand side in real time according to the interruptible load, establish an adjustable power generation resource model at different time scales, and obtain the adjustable energy storage characteristics on the power supply side and the demand side.

[0027] As a preferred embodiment of the present invention, the adjustable power quantity characteristics of the power supply side and the demand side are analyzed through the adjustable power quantity model of the power generation resources, and different power resource characteristics are aggregated, including:

[0028] According to the adjustable energy storage characteristics, the power plant power generation scenarios on the power supply side and the demand side are analyzed, and the equivalent load prediction error between different powers is obtained through the regulation probability distribution;

[0029] It is established that the equivalent load prediction error follows a normal distribution N(μ,σ 2 ) at different time scales, and its expression is:

[0030]

[0031] Among them, μ represents the expected value of the equivalent load prediction error in different time periods, and σ 2 represents the variance of the expected value of the equivalent load prediction error in different time periods, and ΔP l represents the equivalent load power difference in different time periods;

[0032] The equivalent load prediction error replaces other scenarios with similar characteristics through typical power plant power generation scenarios, and the power resource characteristic set under different power plant power generation scenarios is solved.

[0033] As a preferred embodiment of the present invention, based on the different power resource characteristics, a virtual power plant resource aggregation model is established, and the multi-level optimal scheduling data of the virtual power plant is obtained, including:

[0034] Based on the different power resource characteristics, the virtual power plant resources are aggregated and scheduled with the goal of maximizing the overall economic benefit of the virtual power plant, and its expression is:

[0035]

[0036] Among them, β represents the risk preference coefficient in the aggregation process of different power resource characteristics, n represents the time period length at different time scales, θ wi represents the scenario probability under scenario wi, η represents the energy efficiency value, α represents the proportion of the virtual power plant income less than the energy efficiency value η, E pv represents the photovoltaic output income, E w represents the wind turbine output income, E v represents the energy storage device income, E q represents the flexible load income, E r represents the transaction income between the virtual power plant and the power grid;

[0037] The virtual power plant resource data after aggregation scheduling is hierarchically executed to obtain multi-level optimal scheduling data.

[0038] As a preferred embodiment of the present invention, a dynamic peak shaving strategy is adopted to construct an objective function for the multi-level optimized scheduling data, and dynamic constraint conditions are introduced to construct an output scheduling model for the multi-level optimized scheduling data, and real-time power dynamic peak shaving data is obtained, including:

[0039] Taking the minimum curve variance of the equivalent load prediction error at different time scales for the multi-level optimized scheduling data as the objective, where the objective function is:

[0040]

[0041] Where, P li represents the power load value at the i-th moment, P a ′ represents the charge and discharge power of the energy storage at the i-th moment, and P L represents the daily average load;

[0042] Taking the charge and discharge power and the remaining power of the energy storage of the virtual power plant as constraint conditions, initial prediction data is obtained;

[0043] Using a neural network to iteratively process the initial prediction data, planning the output scheduling data of the energy storage, and obtaining the equivalent load prediction error curve after peak shaving and valley filling and the predicted value of the remaining energy of the energy storage;

[0044] Comparing the size relationship between the predicted value of the remaining energy of the energy storage and the true value, and adjusting the peak shaving and valley filling limit through the limit adjustment method to correct the energy storage error;

[0045] Comparing the true load with the predicted load, and correcting the load prediction error by the specific load method to complete the entire peak shaving and valley filling control strategy, and obtaining real-time power dynamic peak shaving data.

[0046] As a preferred embodiment of the present invention, the constraint condition for the charge and discharge power of the energy storage is: during the dynamic peak shaving process of the virtual power plant, the charge and discharge power P a ′ is always less than the rated power P of the energy storage o , and its expression is:

[0047] |P a ′ | ≤ P o ;

[0048] The constraint condition for the remaining power is: during the dynamic peak shaving process of the virtual power plant, the remaining power of the energy storage device shall not be less than the minimum allowable charge and discharge amount of the energy storage, and shall not be greater than the maximum allowable charge and discharge amount of the energy storage.

[0049] The second aspect of the present invention provides a system for a virtual power plant dynamic peak shaving method based on a neural network, including:

[0050] A power data processing module, which is used to process multi-source power data input by a multi-energy generating unit, and obtain power load values at different time scales after processing the multi-source power data;

[0051] A load data prediction module, which predicts the power load status in real time and monitors the load prediction error after dynamic peak shaving through an equivalent load prediction error curve;

[0052] An optimization scheduling module, which is used for the generation and execution of dynamic peak shaving strategies, perceives the operation status of power loads in real time, inputs the short-term prediction points and scenario prediction points of the power plant into the optimization scheduling module, timely corrects the operation conditions of power loads, and completes the adjustment of the real-time power scheduling plan;

[0053] A virtual power plant control module, which controls the charge and discharge power of the power supply side in real time, detects interruptible loads, and conducts power interaction with energy storage devices.

[0054] The present invention has the following beneficial effects compared with the prior art:

[0055] The present invention adopts a dynamic peak shaving strategy. On the basis of short-term power load prediction, a daily scheduling plan for energy storage output is arranged with the aim of using energy storage to help load peak shaving and valley filling. Taking the minimum load variance as the objective function and the charge and discharge power and remaining power of energy storage as the constraints, an energy storage output scheduling model is established to correct the error caused by load prediction, and an energy storage output scheduling scheme is obtained, which can make the load curve smoother and improve the peak shaving effect.

[0056] Considering the randomness of the output of uncertain power sources such as wind power and photovoltaic power and the existence of random variables such as load prediction errors, by analyzing and modeling the flexible adjustability and overall randomness of aggregated elements, the dynamic characteristics of key elements are analyzed, which provides a theoretical support for the aggregator to select the aggregated elements of the virtual power plant. At the level of aggregated regulation characteristics, on the one hand, the external adjustable flexibility of the virtual power plant is obtained by integrating flexible adjustable resources, and on the other hand, the randomness model of the virtual power plant is obtained by operating the probability density function of distributed resources through convolution, ensuring the stability of power grid operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only exemplary, and for those of ordinary skill in the art, other implementation drawings can be obtained by extending according to the provided drawings without creative efforts.

[0058] Figure 1 It is a flowchart of the virtual power plant dynamic peak shaving method provided by the embodiment of the present invention;

[0059] Figure 2 It is a structural block diagram of the virtual power plant dynamic peak shaving system provided by the embodiment of the present invention. Specific embodiments

[0060] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0061] The first embodiment: As Figure 1 shown, the present invention provides a virtual power plant dynamic peak shaving method based on a neural network, including the following steps:

[0062] Obtain the system scheduling instructions of the power supply side and the demand side of the virtual power plant in real time, construct the power generation controllable resource data at different time scales for the system scheduling instructions, analyze the power characteristics of the power generation controllable resource data, and establish the adjustable power quantity model of the power generation resources at different time scales;

[0063] In this embodiment, by obtaining the operation characteristics and aggregation effects of distributed resources in real time, and considering the demands of various energies as a whole, the optimal allocation of various resources is realized at different time scales, and the power generation controllable resource data is obtained.

[0064] Analyze the adjustable power quantity characteristics of the power supply side and the demand side through the adjustable power quantity model of the power generation resources, converge the characteristics of different power resources, establish the virtual power plant resource aggregation model, and obtain the multi-level optimal scheduling data of the virtual power plant;

[0065] In this embodiment, on the power supply side, it can be divided into controllable resources and semi-controllable resources according to the unit controllability; on the load side, based on the two response principles of price-based and incentive-based, it can be further divided into shiftable demand-side resources and interruptible demand-side resources. According to the flexibility of the adjustable resources, the scheduling instructions of the controllable power supply response system and the power fluctuations of the tracking system are aggregated to obtain the multi-level optimal scheduling data of the virtual power plant.

[0066] Adopt a dynamic peak shaving strategy to construct an objective function for the multi-level optimal scheduling data, introduce dynamic constraint conditions to construct an output scheduling model for the multi-level optimal scheduling data, and obtain real-time power dynamic peak shaving data.

[0067] In this embodiment, first, the adjustable flexibility and randomness of different distributed power sources, traditional units, controllable loads, and energy storage are analyzed, and a virtual power plant resource aggregation model of each aggregation element is established at the level of flexible adjustable resources; second, at the level of aggregation regulation characteristics, on the one hand, the adjustable flexibility of the virtual power plant is obtained by integrating the flexible adjustable resources, and on the other hand, the randomness model of the virtual power plant is obtained by operating on the probability density function of the distributed resources through convolution. Finally, an optimization configuration model considering the randomness of the virtual power plant is established based on the portfolio theory, and the internal unit capacities that need to be dynamically matched are analyzed by introducing dynamic constraint conditions, providing a configuration plan for the multi-level optimal scheduling of the internal resources of the virtual power plant.

[0068] Obtain the system scheduling instructions of the power supply side and the demand side of the virtual power plant in real time, and construct the power generation controllable resource data at different time scales for the system scheduling instructions, including:

[0069] Statistical characteristic indicators of the power generation resources corresponding to the system scheduling instructions, and divide the characteristic indicators into the regulation rate of power generation resources, adjustable power, and adjustable power quantity;

[0070] In this embodiment, according to the reliability, stability of the controllable resources and the system scheduling instructions, the characteristic indicators of the power generation resources are obtained, and the characteristic indicators are established as a power generation resource adjustable power quantity diagram with the power axis as the vertical axis and the time axis as the horizontal axis.

[0071] Establish a power generation resource adjustable power quantity diagram with the power axis as the vertical axis and the time axis as the horizontal axis at different time scales according to the characteristic indicators;

[0072] In this embodiment, the area enclosed by the power change curve and the coordinate axis at different moments in the power generation resource adjustable power quantity diagram represents the flexibility interval of the adjustable resources, the curve area enclosed by the maximum power values represents the maximum upward flexibility of the adjustable resources, and the curve area enclosed by the minimum power values represents the minimum downward flexibility of the adjustable resources.

[0073] The power generation resource adjustable power quantity diagram starts from the initial power generation operation point (0, p 0 ), and at any moment t = τ 1 , t = τ 2 , calculate the upward power quantity E 1 and the downward power quantity E 2 respectively, and their expressions are:

[0074]

[0075] Among them, p 1 , p 2 represent the power quantity power values at the moments of τ 1 , τ 2 respectively, respectively represent the maximum adjustable power and the minimum adjustable power during the power adjustment process, p max represents the upper limit value of power adjustment, p min represents the lower limit value of power adjustment, P c represents the regulation efficiency value of power generation resources;

[0076] In this embodiment, the adjustable power of the power generation resources at any moment can be obtained by subtracting the known maximum and minimum output powers of the adjustable resources from the current operating power.

[0077] According to the adjustable power diagram of the power generation resources at τ 1 to τ 2 Within the time scale, regulate the power according to the maximum power operation to obtain the data of the controllable power generation resources at different time scales.

[0078] In this embodiment, in the initial stage of the adjustment of the adjustable power of the power generation resources, that is, when the output is relatively low, the adjustable power of the unit is a constant value, and its flexibility is limited by the regulation rate of the power generation resources; when the output of the unit gradually increases and the upward adjustment space is less than the ramping ability, its flexibility is limited by its own adjustment ability; this flexibility adjustment index reflects the maximum upward and downward adjustable capacity of the unit, and at the same time characterizes the regulation rate of the power generation resources and the rising rate of the power of the power generation resources.

[0079] Analyze the power characteristics of the data of the controllable power generation resources, and establish an adjustable power model of the power generation resources at different time scales, including:

[0080] Divide the data of the controllable power generation resources into M types according to the power generation scenarios of the virtual power plant γ={γ 1 ,γ 2 ,…,γ i ,…,γ M}, i∈(1,M), and take the single power generation data of the γ i type as the data center γ DC ;

[0081] Count the data centers of the M types γ, and set the regulation probability for each data center to obtain the regulation probability distribution of the corresponding controllable power generation resource data, and its expression is:

[0082]

[0083] Among them, D represents the power fluctuation amplitude generated during the power regulation in the power market;

[0084] In this embodiment, according to the different power generation scenarios of the virtual power plant, establish a dynamic peak shaving scenario data center, obtain the hidden information in the operation of the virtual power plant, establish a probability distribution area belonging to the virtual power plant management and control, so as to obtain higher peak shaving profits.

[0085] Maintain the total power conservation of the power supply side and the demand side according to the regulation probability distribution, and respond to the interruptible load on the demand side through the load characteristics during the power time shift process. Its expression is:

[0086] E 1 (Δτ, P L ) = Δτ × min((P es - P e ), (E e - E emax ));

[0087] Among them, Δτ represents the time difference between τ 1 , τ 2 moments, P es represents the maximum charging power included in the controllable power generation resource data, P e represents the power generation power at the current moment, E e the current energy storage power, E emax represents the maximum adjustable power within the time period of τ 1 , τ 2 ;

[0088] Analyze the dispatchable power resources on the demand side in real time according to the interruptible load, establish an adjustable power model of the power generation resources at different time scales, and obtain the adjustable energy storage characteristics of the power supply side and the demand side.

[0089] In this embodiment, the virtual power plant selects the optimal value of the regulation probability distribution corresponding to its hidden information to maximize its own utility, thereby exposing the true hidden information, and solving the problem of information asymmetry between the power supply side and the demand side during the peak shaving process. In the full information scenario where the true information of the virtual power plant can be obtained by the data center, the data center understands the type of each virtual power plant in advance, including the true peak shaving revenue situation of the virtual power plant. In this scenario, when the data center participates in the peak shaving trading process of the virtual power plant, it can retain the utility for it by minimizing the utility of the virtual power plant, thereby obtaining a higher utility than in the incomplete information scenario.

[0090] Analyze the adjustable power characteristics of the power supply side and the demand side through the adjustable power model of the power generation resources, and converge different power resource characteristics, including:

[0091] Analyze the power generation scenarios of the power plants on the power supply side and the demand side according to the adjustable energy storage characteristics, and obtain the equivalent load prediction error between different powers through the regulation probability distribution;

[0092] Establish that the equivalent load prediction error follows a normal distribution N(μ, σ 2 ) at different time scales, and its expression is:

[0093]

[0094] Among them, μ represents the expected value of the equivalent load prediction error in different time periods, and σ 2 represents the variance of the expected value of the equivalent load prediction error in different time periods, and ΔP l represents the power difference of the equivalent load in different time periods;

[0095] The equivalent load prediction error replaces other scenarios with similar characteristics through typical power plant generation scenarios, and solves the power resource feature set under different power plant generation scenarios.

[0096] In this embodiment, the equivalent load prediction error is calculated according to the corresponding regulation probability distribution region.

[0097] Based on different power resource characteristics, a virtual power plant resource aggregation model is established to obtain multi-level optimal scheduling data of the virtual power plant, including:

[0098] Based on different power resource characteristics, with the goal of maximizing the overall economic benefit of the virtual power plant, the resources of the virtual power plant are aggregated and scheduled, and its expression is:

[0099]

[0100] Among them, β represents the risk preference coefficient in the aggregation process of different power resource characteristics, n represents the time period length at different time scales, θ wi represents the scenario probability under scenario wi, η represents the energy efficiency value, α represents the proportion of the virtual power plant's income less than the energy efficiency value η, and E pv represents the photovoltaic output income, E w represents the wind turbine output income, E v represents the energy storage device income, E q represents the flexible load income, E r represents the trading income between the virtual power plant and the power grid;

[0101] The virtual power plant resource data after aggregation and scheduling is hierarchically executed to obtain multi-level optimal scheduling data.

[0102] Adopt a dynamic peak shaving strategy to construct an objective function for the multi-level optimal scheduling data, introduce dynamic constraint conditions to construct an output scheduling model for the multi-level optimal scheduling data, and obtain real-time power dynamic peak shaving data, including:

[0103] With the goal of minimizing the curve variance of the equivalent load prediction error at different time scales for the multi-level optimal scheduling data, the objective function is:

[0104]

[0105] Among them, P li represents the power load value at the i-th moment, P a′ represents the charge and discharge power of energy storage at the i-th moment, P L represents the daily average load;

[0106] In this embodiment, the target function is used to ensure that the remaining power of the energy storage is within a reasonable range, which is beneficial to battery maintenance and convenient for scheduling in case of abnormalities. In addition, the charge and discharge power of the energy storage should also be limited within its rated power range, otherwise the output scheduling cannot be fully executed.

[0107] Taking the charge and discharge power and remaining power of the energy storage in the virtual power plant as constraint conditions, initial prediction data is obtained;

[0108] The initial prediction data is processed iteratively by a neural network to plan the output scheduling data of the energy storage, and the equivalent load prediction error curve after peak shaving and valley filling and the predicted value of the remaining energy of the energy storage are obtained;

[0109] In this embodiment, according to the initial prediction data, considering the maximum output limit of the wind-solar power plant, the remaining power situation and output of the energy storage power station and other constraints, the output of the energy storage power station is scheduled to achieve the best overall economic benefits, the maximum absorption of wind power and photovoltaic power, and the purpose of smoothing the equivalent load prediction error curve.

[0110] Compare the size relationship between the predicted value of the remaining energy of the energy storage and the true value, and adjust the peak shaving and valley filling limit value by the limit adjustment method to correct the energy storage error;

[0111] Compare the true load with the predicted load, and correct the load prediction error by the specific load method to complete the entire peak shaving and valley filling control strategy and obtain real-time power dynamic peak shaving data.

[0112] In this embodiment, according to the load prediction data, the output of the energy storage is scheduled in real time to obtain the equivalent load prediction error curve after peak shaving and valley filling and the predicted value of the remaining energy of the energy storage. Then, compare the actual load with the predicted load. In the case of too large a deviation, reschedule the output of the energy storage. Then, compare the actual remaining energy situation with the predicted remaining energy situation. In the case of too large a deviation, reschedule the output of the energy storage by the limit adjustment method of adjusting the peak shaving and valley filling limit value, starting from both the output of the energy storage and the remaining power of the energy storage to achieve precise peak shaving.

[0113] The constraint condition for the charge and discharge power of the energy storage is: during the dynamic peak shaving process in the virtual power plant, the charge and discharge power P of the energy storage a ′ is always less than the rated power P of the energy storage o , and its expression is:

[0114] |P a ′ | ≤ P o ;

[0115] The remaining power constraint is that during the dynamic peak shaving process of the virtual power plant, the remaining power of the energy storage device shall not be less than the minimum allowable charge and discharge amount of the energy storage and shall not be greater than the maximum allowable charge and discharge amount of the energy storage.

[0116] In this embodiment, the energy storage output scheduling scheme needs to take into account the peak shaving effect on the net load after the addition of the wind-solar-storage system and the economic benefits of users and power plants in order to meet the basic expectations placed on a peak shaving strategy. Three objective functions, namely the minimum net load variance, the minimum electricity cost, and the minimum combined output cost of wind-solar-storage, are selected for multi-objective optimization. The smaller the net load variance, the flatter the net load curve, and the better the peak shaving effect of the wind-solar-storage combined system. Therefore, the minimum variance of the net load after combined peak shaving and valley filling is taken as the objective function.

[0117] The second embodiment: As Figure 2 shown, a system for a virtual power plant dynamic peak shaving method based on a neural network includes:

[0118] A power data processing module for processing multi-source power data input by multi-energy generating units and obtaining power load values at different time scales after processing the multi-source power data;

[0119] A load data prediction module for predicting the power load status in real time and monitoring the load prediction error after dynamic peak shaving through an equivalent load prediction error curve;

[0120] An optimization scheduling module for generating and executing dynamic peak shaving strategies, perceiving the operation trend of the power load in real time, inputting the short-term prediction points and scenario prediction points of the power plant into the optimization scheduling module, and timely correcting the operation conditions of the power load to complete the adjustment of the real-time power scheduling plan;

[0121] A virtual power plant control module for controlling the charge and discharge power of the power supply side in real time, detecting interruptible loads, and performing power interaction with energy storage devices.

[0122] In this embodiment, the virtual power plant control module receives the scheduling instructions of the optimization scheduling module, and then the load data prediction module combines the data of the load prediction system and the requirements of the wind-solar-storage combined scheduling, and issues charge and discharge instructions after calculation, which can not only improve the wind-solar energy consumption ratio but also reduce costs, with good economic benefits.

[0123] The present invention adopts a dynamic peak shaving strategy. Based on the short-term power load prediction, a day-ahead scheduling plan for the energy storage output is arranged with the aim of using the energy storage to help the load shave peaks and fill valleys. The minimum load variance is used as the objective function, and the energy storage charge and discharge power and the remaining power are used as constraints to establish an energy storage output scheduling model, correct the error caused by the load prediction, and obtain an energy storage output scheduling scheme, which can make the load curve smoother and improve the peak shaving effect.

[0124] Considering the randomness of the output of uncertain power sources such as wind power and photovoltaic power, as well as the existence of random variables such as load forecasting errors, by analyzing and modeling the flexible adjustability and overall randomness of aggregated elements, the dynamic characteristics of key elements are analyzed, providing theoretical support for the aggregator to select the aggregated elements of the virtual power plant. At the level of aggregated regulation characteristics, on the one hand, the external adjustable flexibility of the virtual power plant is obtained by integrating flexible adjustable resources, and on the other hand, the randomness model of the virtual power plant is obtained by operating on the probability density function of distributed resources through convolution, ensuring the stability of power grid operation.

[0125] The above embodiments are only exemplary embodiments of the present application and are not used to limit the present application. The protection scope of the present application is defined by the claims. Those skilled in the art can make various modifications or equivalent replacements within the essence and protection scope of the present application, and such modifications or equivalent replacements should also be regarded as falling within the protection scope of the present application.

Claims

1. A virtual power plant dynamic peak regulation method based on neural network, characterized in that: The following steps are involved: Acquire system dispatch instructions on the power supply side and demand side of the virtual power plant in real time, construct controllable power generation resource data at different time scales for the system dispatch instructions, analyze the power characteristics of the controllable power generation resource data, and establish an adjustable power model of power generation resources at different time scales; Analyze the adjustable power characteristics of the power supply and demand side through the adjustable power model of the power generation resources, aggregate the characteristics of different power resources, establish a virtual power plant resource aggregation model, and obtain the multi-level optimization scheduling data of the virtual power plant; Adopting a dynamic peak-shaving strategy to construct an objective function for the multi-level optimization scheduling data, introducing dynamic constraint conditions to construct an output scheduling model for the multi-level optimization scheduling data, and obtaining real-time power dynamic peak-shaving data; According to the characteristics of the different power resources, a virtual power plant resource aggregation model is established to obtain multi-level optimization scheduling data of the virtual power plant, including: According to the characteristics of different power resources, the virtual power plant resources are aggregated and dispatched with the goal of maximizing the overall economic benefits of the virtual power plant. The expression is: Among them, β represents the risk preference coefficient of different power resource characteristics in the aggregation process, n represents the length of the time period at different time scales, represents the scenario probability under scenario wi, η represents the energy efficiency value, α represents the proportion of virtual power plant benefits less than the energy efficiency value η, E pv Represents the photovoltaic output income, E w Indicates the wind turbine output benefit, E v represents the benefit of energy storage equipment, E q represents the flexible load benefit, E r represents the transaction revenue between the virtual power plant and the grid; The aggregated dispatched virtual power plant resource data is hierarchically executed to obtain multi-level optimized dispatch data; A dynamic peak-shaving strategy is adopted to construct an objective function for the multi-level optimization dispatching data, and a dynamic constraint condition is introduced to construct an output dispatching model for the multi-level optimization dispatching data to obtain real-time power dynamic peak-shaving data, including: The multi-level optimization dispatching data aims to minimize the curve variance of the equivalent load forecast error at different time scales, where the objective function is: Among them, P li represents the power load value at the i-th moment, P a ′ represents the energy storage charging and discharging power at the i-th moment, P L Indicates the average load of the day; Using the energy storage charging and discharging power and remaining power of the virtual power plant as constraints, initial forecast data is obtained; Iteratively process the initial prediction data using a neural network, plan the energy storage output dispatch data, and obtain an equivalent load prediction error curve after peak shaving and valley filling and a predicted value of the remaining energy of the energy storage; Compare the relationship between the predicted value of energy storage remaining energy and the actual value, and adjust the peak shaving and valley filling limit value through the limit adjustment method to correct the energy storage error; Compare the actual load with the predicted load, use the load ratio method to correct the load prediction error, complete the entire peak shaving and valley filling control strategy, and obtain real-time dynamic power peak regulation data.

2. According to the neural network-based virtual power plant dynamic peak-shaving method of claim 1, it is characterized in that: Obtain system dispatch instructions on the power supply side and demand side of the virtual power plant in real time, and construct controllable power generation resource data at different time scales for the system dispatch instructions, including: Counting characteristic indicators of power generation resources corresponding to the system scheduling instructions, and dividing the characteristic indicators into power generation resource adjustment rate, adjustable power and adjustable power; According to the characteristic index, an adjustable power diagram of power generation resources is established at different time scales, with the vertical axis being the power axis and the horizontal axis being the time axis; The adjustable power diagram of power generation resources takes the initial power generation operation point (0, p0) as the starting point, and calculates the upward power E1 and downward power E2 at any time t=τ1 and t=τ2 respectively, and its expression is: Among them, p1 and p2 represent the power values ​​at time τ1 and τ2 respectively. They represent the maximum and minimum adjustable power values ​​during the regulation process, respectively. max Indicates the upper limit of power regulation, p min Indicates the lower limit of power regulation, P c Indicates the regulation efficiency value of power generation resources; According to the adjustable power diagram of power generation resources, within the time scale of τ1 to τ2, the power is regulated according to the maximum power operation to obtain the controllable power generation resource data at different time scales.

3. The method for dynamic peak load regulation of a virtual power plant based on a neural network according to claim 2, characterized in that: Analyze the power characteristics of the controllable power resource data and establish an adjustable power model of power resources at different time scales, including: The controllable power generation resource data is divided into M types according to the virtual power plant power generation scenario γ = {γ1,γ2,…,γ i ,…,γ M },i∈(1,M), and γ i Type of single generation data as data center γ DC ; M data centers of type γ are counted, and a control probability is set for each data center to obtain the control probability distribution corresponding to the power generation controllable resource data, which is expressed as: Where D represents the power fluctuation amplitude generated when power regulation occurs in the power market; According to the control probability distribution, the total power conservation on the power supply side and the demand side is maintained, and the interruptible load on the demand side is dealt with by the load characteristics during the power time shift process. The expression is: E1(Δτ,P L )=Δτ×min((P es -P e ),(THE e -THE emax )); Among them, Δτ represents the time difference between τ1 and τ2, P es Indicates the maximum charging power contained in the controllable resource data of power generation, P e Indicates the power generated at the current moment, E e Current energy storage capacity, E emax Indicates the maximum adjustable power in the time period of τ1 and τ2; According to the interruptible load, the dispatchable power resources on the demand side are analyzed in real time, an adjustable power model of power generation resources under different time scales is established, and the adjustable energy storage characteristics of the power supply and demand side are obtained.

4. The method for dynamic peak load regulation of a virtual power plant based on a neural network according to claim 3 is characterized in that: The adjustable power model of power generation resources is used to analyze the adjustable power characteristics of power supply and demand side, and to gather the characteristics of different power resources, including: Analyze the power generation scenario of the power plant on the power supply side and the demand side according to the adjustable energy storage characteristics, and obtain the equivalent load prediction error between different power sources through the regulation probability distribution; Establish that the equivalent load forecast error obeys N(μ,σ 2 ) is normally distributed, and its expression is: Where μ represents the expected value of the equivalent load forecast error in different time periods, σ 2 The variance of the expected value of the equivalent load forecast error in different time periods, ΔP l Indicates the power difference of equivalent load in different time periods; The equivalent load prediction error replaces other scenarios with similar characteristics by a typical power plant power generation scenario, and solves the power resource feature set under different power plant power generation scenarios.

5. The method for dynamic peak load regulation of a virtual power plant based on a neural network according to claim 1, characterized in that: The energy storage charging and discharging power constraint condition is: during the dynamic peak regulation process of the virtual power plant, the energy storage charging and discharging power P a ′ Always less than the energy storage rated power P o , whose expression is: |P a ′ |≤P o ; The remaining power constraint condition is: during the dynamic peak regulation of the virtual power plant, the remaining power of the energy storage device shall not be less than the minimum allowable charge and discharge capacity of the energy storage, and shall not be greater than the maximum allowable charge and discharge capacity of the energy storage.

6. A virtual power plant dynamic peak-shaving system based on a neural network based on the virtual power plant dynamic peak-shaving method based on a neural network according to any one of claims 1 to 5, characterized in that: include: A power data processing module is used to process multi-source power data input by multi-energy generators, and obtain power load values ​​at different time scales after processing the multi-source power data; The load data prediction module predicts the power load status in real time and monitors the load prediction error after dynamic peak regulation in real time through the equivalent load prediction error curve; The optimization scheduling module is used to generate and execute dynamic peak load regulation strategies, perceive the power load operation status in real time, input the short-term prediction points and scenario prediction points of the power plant into the optimization scheduling module, timely correct the power load operation conditions, and complete the real-time power scheduling plan adjustment; The virtual power plant control module controls the charging and discharging power on the power supply side in real time, detects interruptible loads, and interacts with energy storage devices for power.

Images