Power system optimization method and system based on multi-load demand response incentive

Abstract

The present invention provides a method and system for optimizing an electric power system based on multi-load demand response incentives, which belongs to the technical field of electric power system optimization. The scheme can accurately quantify the randomness and uncertainty of user behavior by describing user portraits of flexible resources on the user side of the electric power system. Based on the obtained user portraits, combined with the demand response mechanism based on utility functions, weighted graphs and shortest paths proposed in the scheme, the accuracy of electric power system optimization is effectively improved. Secondly, the scheme fully considers the randomness and uncertainty of user behavior in the electric power system during the user portrait construction process. By adopting a scheme that combines scenario generation and scenario reduction, the representativeness of the user portrait is guaranteed, and the computational complexity of subsequent optimization solutions is reduced. It provides an accurate and efficient user behavior description for the subsequent demand response mechanism design, thereby effectively improving the efficiency and accuracy of electric power system optimization.

Description

Technical Field

[0001] The present invention belongs to the technical field of power system optimization, and in particular relates to a power system optimization method and system based on multi-load demand response incentive. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] As an important means to optimize the operation of power systems and improve energy efficiency, the mechanism design of demand response involves multiple technical fields and theoretical foundations. With the continuous increase in the penetration rate of renewable energy, the volatility and uncertainty of the power system have become increasingly prominent. Traditional coping solutions such as expanding the capacity of generator sets face the problem of diminishing marginal benefits, and the regulation capacity of the power generation side is becoming increasingly limited. In this context, formulating a reasonable demand response incentive mechanism has become an effective alternative. The demand response incentive mechanism forms a virtual distributed energy resource pool by coordinating and optimizing the adjustable loads on the user side, such as smart homes, electric vehicle charging, and interruptible industrial and commercial loads. It participates in power market transactions and power grid dispatching as a whole, and provides auxiliary services such as peak shaving, frequency regulation, and voltage support, thereby improving the flexibility and stability of the power grid. The design of the demand response incentive mechanism needs to comprehensively consider multiple factors such as user behavior characteristics, load characteristics, market mechanisms, and system operation. It is a key technology to achieve power supply and demand balance and improve energy efficiency.

[0004] At present, the commonly used power system optimization scheme mainly adopts demand response based on VCG (Vickrey-Clarke-Groves) mechanism. The VCG mechanism achieves incentive compatibility through clever payment rules, but the VCG mechanism is computationally complex. As the number of participants increases, the computational complexity of the VCG mechanism increases exponentially, making it difficult to apply to large-scale power systems. At the same time, the traditional VCG mechanism design method does not adequately model user behavior and does not fully consider the randomness and uncertainty of user behavior, making it difficult to adapt to the rapid changes and diversified demands of the power market. Summary of the invention

[0005] The present invention provides a method and system for power system optimization based on multi-load demand response incentives to solve the problem that the traditional demand response incentive mechanism lacks consideration of the randomness and uncertainty of user behavior, resulting in insufficient accuracy of power system optimization and difficulty in application to large-scale power systems.

[0006] According to a first aspect of an embodiment of the present invention, a method for optimizing a power system based on multi-load demand response incentive is provided, comprising:

[0007] Perform scenario analysis on the flexible resources on the user side of the power system to generate user portraits corresponding to the flexible resources; wherein the flexible resources include distributed power sources and flexible loads;

[0008] For each user profile of the flexible resource, construct the corresponding utility function;

[0009] Based on the constructed utility function, the power system optimization objective function is constructed with the goal of maximizing the difference between the utility of all flexible resources and the total energy cost;

[0010] Based on the user portraits corresponding to the flexible resources, a weighted graph is constructed for each flexible resource;

[0011] Based on the weighted graph of each flexible resource, the feasible mode for each flexible resource to participate in demand response is determined by calculating the shortest path from the auxiliary node to any user portrait node;

[0012] Based on the feasible modes corresponding to all flexible resources, a solution set that meets the current power system needs is constructed;

[0013] Based on the obtained solution set, the optimal solution participating in the power system optimization is obtained from the solution set by maximizing the power system optimization objective function.

[0014] Furthermore, the generation of the user portrait is specifically as follows: for distributed power sources, based on the historical grid-connected power data of distributed power sources, a pre-trained grid-connected power scenario generation model based on deep learning is used to obtain several distributed power source grid-connected scenarios; for flexible loads, based on the historical demand response rate of flexible loads, a Monte Carlo simulation is used to obtain several flexible load response rate scenarios; for the obtained several distributed power source grid-connected scenarios and several flexible load response rate scenarios, a pre-trained variational encoder is used to reduce the scenarios, and the reduced scenarios are used as user portraits of distributed power sources and flexible loads.

[0015] Furthermore, the utility function is used to quantify the utility values ​​corresponding to different user profiles of flexible resources, and to quantify the utility values ​​into price indicators by introducing standard electricity prices, wherein utility parameters are introduced into the utility function, and the utility parameters are determined based on the priority coefficient of the flexible resources participating in demand response, the standard electricity price, and the product of the response characteristic function.

[0016] Furthermore, the utility function is specifically expressed as follows:

[0017]

[0018] in, Represents flexible resources The corresponding utility value; Represents flexible resources The amount of electricity or electricity demand in the grid during period t, is the utility parameter, which is expressed as the product of the priority coefficient of the flexible resource participating in the demand response, the standard electricity price and the response characteristic function; To adjust the parameters.

[0019] Furthermore, the weighted graph uses different types of user portraits corresponding to flexible resources as nodes and introduces auxiliary nodes; edges are constructed based on whether type conversion can be performed between nodes, and the weight of the edge represents the utility change caused by the conversion between flexible resource nodes.

[0020] Furthermore, the construction of the weighted graph is specifically as follows: different types of user portraits corresponding to flexible resources are used as nodes, and auxiliary nodes are introduced; directed edges are set from the auxiliary nodes to other user portrait nodes, and whether directed edges are set between user portrait nodes is determined based on whether type conversion can be performed; each directed edge is set with a weight for indicating the length of the edge, and the weight is determined based on the utility values ​​corresponding to the two nodes.

[0021] Furthermore, the weight is determined as follows: for any first node and second node, the weight of the directed edge from the first node to the second node is the utility value of the second node minus the utility value of the first node; the weight of the directed edge from the second node to the first node is the utility value of the first node minus the utility value of the second node.

[0022] Furthermore, the shortest path from the auxiliary node to any user portrait node aims to solve the maximum utility path from the auxiliary node to any portrait node, and the shortest path is solved using the Dijkstra algorithm.

[0023] Furthermore, the power system optimization objective function is specifically expressed as follows:

[0024]

[0025] in, Represents flexible resources The corresponding utility value, is the total energy cost during period t, is the total load during period t.

[0026] According to a second aspect of an embodiment of the present invention, there is provided a power system optimization system based on multi-load demand response incentive, comprising:

[0027] A user portrait generation unit, which is used to perform scenario analysis on the flexible resources on the user side of the power system and generate a user portrait corresponding to the flexible resources; wherein the flexible resources include distributed power sources and flexible loads;

[0028] A utility function construction unit, which is used to construct a corresponding utility function for each user profile of the flexible resource;

[0029] An objective function construction unit, which is used to construct an objective function for optimizing the power system based on the constructed utility function with the goal of maximizing the difference between the utility of all flexible resources and the total energy cost;

[0030] A weighted graph construction unit, which is used to construct a weighted graph for each flexible resource based on the user profile corresponding to the flexible resource;

[0031] A solution set construction unit is used to determine the feasible mode for each flexible resource to participate in demand response based on the weighted graph of each flexible resource by calculating the shortest path from the auxiliary node to any user portrait node; based on the feasible modes corresponding to all flexible resources, a solution set that meets the current power system demand is constructed;

[0032] The optimization solving unit is used to obtain the optimal solution for the power system optimization from the solution set by maximizing the power system optimization objective function based on the obtained solution set.

[0033] One or more of the above technical solutions have the following beneficial effects:

[0034] The present invention provides an electric power system optimization system based on multi-load demand response incentives. The scheme can accurately quantify the randomness and uncertainty of user behavior by describing the user profile of the flexible resources on the user side of the electric power system. Based on the obtained user profile, combined with the demand response mechanism based on utility function, weighted graph and shortest path proposed in the scheme, the accuracy of electric power system optimization is effectively improved.

[0035] The present invention adopts a scheme combining scenario generation and scenario reduction in the process of constructing user portraits. This scheme fully considers the randomness and uncertainty of user behavior in the power system: in the scenario generation stage, a time series analysis method based on deep learning is adopted for distributed power sources, which can effectively capture the periodicity and seasonal characteristics of electricity data, and a Monte Carlo simulation method is adopted for flexible loads to simulate the randomness of user response behavior; in the scenario reduction stage, a large number of random scenarios are converted into typical scenarios under certain probabilities through variational autoencoders (i.e., VAE). The variational autoencoder can not only automatically learn the important features of the data and reduce artificial feature engineering, but also generate new reasonable scenarios according to specific conditions. This scheme combining scenario generation and scenario reduction not only ensures the representativeness of user portraits, but also reduces the computational complexity of subsequent optimization solutions, providing accurate and efficient user behavior description for the subsequent demand response mechanism design.

[0036] The solution described in the present invention constructs a demand response mechanism based on weighted graphs and shortest paths, proposes a graph construction method in which nodes represent user portrait types and edge weights reflect user valuations, and solves the shortest path based on the Dijkstra algorithm to obtain a feasible mode for users. This method effectively coordinates the interests of multiple types of users, and the incentive mechanism has good flexibility, which can effectively improve the overall utility of power system optimization.

[0037] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0039] Figure 1 It is a flow chart of the STL (Seasonal and Trend decomposition using Loess) algorithm described in an embodiment of the present invention;

[0040] Figure 2 This is an example of a weighted graph described in an embodiment of the present invention.

[0041] Figure 3 Flow chart of the Dijkstra algorithm described in an embodiment of the present invention.

[0042] Figure 4 The present invention is a flowchart of a method for optimizing a power system based on multiple load demand response incentives described in an embodiment of the present invention. DETAILED DESCRIPTION

[0043] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0044] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0045] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0046] Terminology explanation:

[0047] Flexible resources: refers to power resources in the power system that have controllable and adjustable characteristics and can actively change their power generation / power consumption behavior according to system needs. They have the following characteristics: real-time or quasi-real-time adjustment capabilities; can participate in the power market and system dispatch; and can provide auxiliary services.

[0048] In the present invention, flexible resources specifically refer to two types of resources on the user side:

[0049] 1) Distributed power sources, including but not limited to distributed photovoltaics, small wind turbines, and energy storage systems (such as electric vehicles with vehicle-to-grid functions). These power sources can participate in demand response by adjusting the amount of electricity connected to the grid;

[0050] 2) Flexible loads, including but not limited to electric vehicle charging loads, HVAC and hot water systems, etc. These loads can participate in demand response by adjusting the power consumption period and power consumption.

[0051] The above-mentioned flexible resources form a virtual distributed energy resource pool through the multiple load demand response incentive mechanism proposed in the present invention, providing auxiliary services such as peak shaving, frequency regulation and voltage support for the power grid.

[0052] In one or more embodiments, Figure 4 As shown, this embodiment provides a power system optimization method based on multi-load demand response incentives, including:

[0053] Step 1: Perform scenario analysis on the flexible resources on the user side of the power system to generate a user profile corresponding to the flexible resources; wherein the flexible resources include distributed power sources and flexible loads;

[0054] It can be understood that the distributed power sources described in this embodiment include but are not limited to distributed photovoltaics, small wind turbines, and energy storage systems (for example, electric vehicles, which can be used as an energy storage system to supply power to the power grid); the flexible loads include but are not limited to electric vehicles, HVAC and hot water systems.

[0055] In a specific implementation, the user portrait is generated as follows: for distributed power sources, based on the historical grid-connected power data of distributed power sources, a pre-trained grid-connected power scenario generation model based on deep learning is used to obtain several distributed power source grid-connected scenarios; for flexible loads, based on the historical demand response rate of flexible loads, a Monte Carlo simulation is used to obtain several flexible load response rate scenarios; for the obtained several distributed power source grid-connected scenarios and several flexible load response rate scenarios, a pre-trained variational encoder is used to reduce the scenarios, and the reduced scenarios are used as user portraits of distributed power sources and flexible loads.

[0056] Specifically, for distributed power sources and flexible loads, the generation of user portraits includes the following processing steps:

[0057] In the scheme described in this embodiment, in view of the randomness of the power of the flexible resources on the user side and the uncertainty of the user behavior, the scenario analysis method is used to establish user portraits of different flexible resources. With the introduction of flexible resources such as distributed power sources and flexible loads, the power on the user side is affected by multiple factors such as weather and user behavior, and the randomness increases. The user's willingness to respond is also closely related to people's complex behavior and psychological activities, and has a certain degree of uncertainty. The scenario analysis method is an effective method to solve the randomness and uncertainty of data. Through scenario generation and scenario reduction, the uncertain random scenarios are converted into different clusters under certain probabilities, thereby generating user portraits of flexible resources on the user side.

[0058] Step 101: scene generation, which specifically includes:

[0059] The time series analysis method is used to generate the grid-connected power of flexible resources such as distributed power sources and flexible loads, and the Monte Carlo simulation method is used to generate the response rate of flexible loads such as air conditioners and electric vehicles for each household.

[0060] In this embodiment, time series analysis and Monte Carlo simulation are mainly used to achieve scenario generation, where time series analysis is a data prediction method that can capture the periodic and seasonal trends of time series data. Monte Carlo simulation is a mathematical method based on random numbers and is widely used to simulate complex systems with randomness and uncertainty.

[0061] Specifically, for distributed power sources, taking distributed photovoltaic and electric vehicles as examples, the scenario generation of distributed power sources is explained, which specifically includes the following process:

[0062] (1) Collect historical data sets of distributed photovoltaic and electric vehicle grid-connected electricity, perform data preprocessing, remove outliers, fill in missing values, and record the processed data sets as (represents the historical data set of distributed photovoltaic grid-connected electricity) and (represents the historical data set of electric vehicle grid-connected electricity);

[0063] (2) Use the STL (Seasonal-Trend decomposition procedure based on Loess) algorithm to decompose the time series. The STL algorithm flow chart is as follows: Figure 1 As shown, the preprocessed data , Decomposition is divided into three parts: trend part, seasonal part and residual part.

[0064] (3) Model selection and training. Pattern recognition is performed on the three parts of data obtained above. Sunrise and sunset and winter and summer are identified in the historical data set of distributed photovoltaic grid-connected electricity. In the historical data set of electric vehicle grid-connected electricity, intraday charging mode, midweek charging mode and weekend charging mode are identified. The LSTM neural network is then used for model training. The parameters of the selected LSTM neural network components are shown in Table 1.

[0065] Table 1 LSTM model component parameters

[0066]

[0067] (4) Generation of grid-connected electricity scenarios. Use the trained LSTM model to generate several possible grid-connected scenarios for distributed photovoltaics and electric vehicles.

[0068] LSTM model input data: Time features: including time of day (24 hours), date (month, season), weather conditions and other time series information; Historical data: historical grid-connected power data of distributed photovoltaics and electric vehicles.

[0069] LSTM model output data: predicted grid-connected power time series, including three components: trend, seasonality, and residual.

[0070] Here are a few examples of grid access scenarios: For distributed photovoltaics, it can be divided into summer sunny scenes: power generation starts at sunrise (6:00), reaches a peak at 12:00, power generation gradually decreases after 16:00, and basically no output after 19:00; winter cloudy scenes: slowly climb after sunrise (7:30), the peak period (12:00-14:00) power generation is lower than in summer, with greater volatility, and changes in cloud cover cause output to fluctuate. For electric vehicles, it can be divided into weekday scenes: the charging demand is high during the morning peak (8:00-10:00), and the charging demand is the largest during the evening peak (18:00-22:00); weekend scenes: the charging demand is more evenly distributed, with two smaller peaks at noon (11:00-14:00) and evening (19:00-23:00).

[0071] The above process aims at the randomness of photovoltaic power generation and electric vehicle grid-connected electricity, and establishes a time series model to generate scenarios. The following uses the Monte Carlo method to generate scenarios for the uncertainty of user demand response. Specifically, for flexible loads, taking air conditioning load and electric vehicle load as examples, the generation of their user portraits is explained, which specifically includes the following processing procedures:

[0072] (1) Based on historical data and expert experience, the response rate of flexible loads such as air conditioning loads and electric vehicles is determined to meet the probability distribution, which is denoted as ,in, is the index of the flexible load type; are distribution parameters estimated based on historical data.

[0073] (2) Identify the influencing factors.

[0074] The key factors that affect the user response rate are determined and expressed with symbols to facilitate subsequent analysis, as shown in Table 2.

[0075] Table 2 Key factors affecting user response rate

[0076]

[0077] Establish a response rate model. Assume that the user response rate follows the Beta distribution. The probability density function expression of the Beta distribution is:

[0078] (1)

[0079] in, is the Beta function, defined as:

[0080] (2)

[0081] in, Represents the gamma function.

[0082] Map influencing factors to distribution parameters.

[0083] (3)

[0084] (4)

[0085] in, and is a mapping function, which is obtained by fitting specific data.

[0086] Set simulation parameters. Determine the number of simulations , define the time range and other relevant parameters.

[0087] Perform a Monte Carlo simulation. For each simulation =1 to , calculate the distribution parameters based on the current input and , thus obtaining the distribution function of user response rate , and then use the distribution function to generate the user response rate.

[0088] Based on specific examples, the results of the air-conditioning load response rate for residential users are as follows: high temperature scenario on summer weekdays: response period: 14:00-16:00, electricity price level: 1.2 yuan / kWh, outdoor temperature: 35°C, expected response rate: 65%-75%; general weather scenario on winter weekends: response period: 19:00-21:00, electricity price level: 0.8 yuan / kWh, outdoor temperature: 5°C, expected response rate: 40%-50%.

[0089] Based on specific examples, the electric vehicle load response rate generation results are as follows: weekday electricity price peak period scenario: response period: 18:00-20:00, electricity price level: 1.5 yuan / kWh, charging demand: high, expected response rate: 80%-90%; weekend electricity price low period scenario: response period: 23:00-5:00 the next day, electricity price level: 0.4 yuan / kWh, charging demand: medium, expected response rate: 20%-30%.

[0090] Step 102: Scene reduction.

[0091] According to the scene generation result of step 101, a variational autoencoder (VAE) is used to perform scene reduction, and the obtained large amount of scene data is converted into deterministic scenes under different probabilities.

[0092] It should be noted that VAE is a generative model that combines deep learning and variational inference. It can capture complex nonlinear relationships in data, provide probabilistic interpretation of data, and better represent uncertainty. It can not only select existing scenarios, but also generate new and reasonable scenarios. It can design conditional VAE to generate scenarios based on specific conditions (such as extreme weather conditions). It can automatically learn important features of data and reduce artificial feature engineering. Therefore, VAE is well suited for scenario reduction problems under complex time series data.

[0093] Specifically, scene reduction includes the following processing steps:

[0094] (1) Data preprocessing

[0095] Convert distributed generation, flexible load power data, user response rate data and related characteristics (time, weather, etc.) into a format suitable for VAE input.

[0096] First, data collection is performed. According to the result of step 101, distributed power supply data is collected: , flexible load: , user response rate data: ,Time characteristics: , weather characteristics: .

[0097] Next, the data is preprocessed and the three types of data, distributed power supply, flexible load and user response rate, are standardized respectively:

[0098] (5)

[0099] in, is the mean, is the standard deviation, represents the data to be standardized, Indicates the result after standardization.

[0100] Convert time to continuous features using periodic encoding:

[0101] (6)

[0102] (7)

[0103] That is, convert each hour of the day into a pair of sine values and cosine .

[0104] And, use one-hot encoding for weather data.

[0105] Finally, all the features after preprocessing are combined into an input vector:

[0106] (8)

[0107] (2) Designing VAE architecture

[0108] The VAE architecture consists of three parts: encoder, decoder, and latent space. Encoder: maps input data to latent space; Decoder: reconstructs original data from latent space; Latent space: low-dimensional space used to capture the core features of data.

[0109] Among them, the design of the encoder includes input layer, hidden layer, mean layer, and log variance layer.

[0110] Input Layer:

[0111] (9)

[0112] in, For input data, R is the field of real numbers, is the input dimension.

[0113] Hidden Layer:

[0114] (10)

[0115] in, h represents the output of the hidden layer, represents the weight matrix of the encoder, Represents the bias vector of the encoder. ReLU() is a type of activation function.

[0116] Mean layer:

[0117] (11)

[0118] Among them, μ represents the mean vector of the latent space, represents the weight matrix of the mean layer, Represents the bias vector of the mean layer.

[0119] Logvariance layer:

[0120] (12)

[0121] in, represents the variance of the latent space, represents the weight matrix of the variance layer, Represents the bias vector of the variance layer.

[0122] The design of the latent space includes the dimension and sampling method.

[0123] Dimensions:

[0124] (13)

[0125] Among them, z represents the latent vector, represents the dimension of the latent space, usually, .

[0126] Sampling method:

[0127] (14)

[0128] in, , represents normally distributed random noise, Represents element-wise multiplication.

[0129] The decoder design includes input layer, hidden layer, and output layer. Input:

[0130] (15)

[0131] in, represents the output of the decoder hidden layer.

[0132] Hidden Layer:

[0133] (16)

[0134] in, represents the weight matrix of the decoder, Represents the bias vector of the decoder.

[0135] Output layer:

[0136] (17)

[0137] in, represents the reconstructed output, represents the weight matrix of the output layer, Represents the bias vector of the output layer, and sigmoid is the activation function of the output layer.

[0138] (3) Train VAE.

[0139] Use historical data to train VAE and optimize reconstruction error and KL divergence.

[0140] Define the loss function:

[0141] (18)

[0142] in:

[0143] (19)

[0144] (20)

[0145] Among them, L is the total loss function, is the reconstruction loss function, is the KL divergence loss. represents the probability distribution of the original data x given the latent variable z, E[ ] represents the expected value, μ is the mean of the encoder output, is the variance of the encoder output.

[0146] During optimization, the loss function is minimized using stochastic gradient descent or one of its variants such as Adam:

[0147] (twenty one)

[0148] in, are the updated parameters of the model, are the current parameters of the model, is the learning rate, is the loss function with respect to the parameters gradient.

[0149] (4) Scene Coding

[0150] Use the trained encoder to map the original scene to the latent space, and for each original scene, calculate its representation in the latent space.

[0151] (5) Representation of clustering latent space

[0152] The K-means algorithm is used to cluster the encoded scenes in the latent space. Specifically:

[0153] initialization Center points: , using formula (22), assign each point to the nearest center, and use formula (23) to recalculate the center of each cluster and update it. Continuously iterate using formula (22) and formula (23) until convergence.

[0154] (twenty two)

[0155] (twenty three)

[0156] in, Represents the i-th element in a cluster (i.e. the original scenario above).

[0157] (6) Select representative scenarios.

[0158] Select the center point from each cluster, specifically:

[0159] For each cluster , from formula (23), we can see that its center is , the point closest to the center is:

[0160] (twenty four)

[0161] (7) Scene reconstruction.

[0162] Use the VAE decoder to reconstruct the selected latent representation into the original space scene. Specifically: for each selected representative latent representation , use the decoder to reconstruct the scene, and the final scene reduction result is ,in, is the number of clusters chosen.

[0163] Step 2: For each user profile of the flexible resource, construct the corresponding utility function;

[0164] In a specific implementation, the utility function is used to quantify the utility values ​​corresponding to different user profiles of flexible resources, and to quantify the utility values ​​into price indicators by introducing standard electricity prices, wherein utility parameters are introduced into the utility function, and the utility parameters are determined based on the priority coefficient of the flexible resources participating in demand response, the standard electricity price, and the product of the response characteristic function.

[0165] In a specific implementation, the priority coefficient of the flexible resource participating in demand response is determined based on the historical data of the flexible resource participating in demand response. Specifically, for a certain flexible resource, based on its historical data, the proportion of the number of demand responses actually completed by the flexible resource in the total number of demand responses it participated in is calculated, and the priority coefficient of the current flexible resource is determined based on the proportion.

[0166] In a specific implementation, the utility function is specifically expressed as follows:

[0167] (twenty three)

[0168] in, Represents flexible resources The corresponding utility value; Represents flexible resources The amount of electricity or electricity demand in the grid during period t, To adjust the parameters, is the utility parameter.

[0169] Specifically, the utility parameter It is expressed as the product of the priority coefficient of flexible resources participating in demand response, the standard electricity price and the response characteristic function.

[0170] In the specific implementation, for distributed power sources, the response characteristic function takes a fixed value; for flexible loads, the response characteristic function is related to the expected value of its response rate. When the response rate is higher than the benchmark value (i.e., the benchmark grid-connected power of the distributed power source or the benchmark power demand of the flexible load), it shows a linear relationship, and when it is lower than the benchmark value, it shows an exponential decay relationship.

[0171] (25)

[0172] Among them, p is the electricity price (yuan / kWh), is the priority coefficient of flexible resource i (dimensionless), is the influence function of the response rate, and its specific form is:

[0173] For flexible loads:

[0174] (26)

[0175] For distributed power generation:

[0176] (27)

[0177] in, is the expected value of the response rate of flexible resource i (obtained from Beta distribution), is the response rate benchmark value, which is 0.5. , , , λ are unknown coefficients, which can be determined by empirical values, where Ensure that the function is continuous.

[0178] Step 3: Based on the constructed utility function, the power system optimization objective function is constructed with the goal of maximizing the difference between the utility of all flexible resources and the total energy cost;

[0179] In the specific implementation, the system model needs to be defined first, specifically:

[0180] For flexible resources i , whose regulation boundary at time t is set to , where for distributed power i , Indicates the minimum adjustable grid-connected power. Indicates the maximum adjustable grid-connected power; for flexible loads i , for flexible loads i : Indicates the minimum power demand Indicates the maximum power demand.

[0181] Next, a convex quadratic function is used to set the energy cost model:

[0182] (twenty four)

[0183] in, for The total load of the time period, To adjust the parameters.

[0184] Furthermore, the power system optimization objective function is specifically expressed as follows:

[0185]

[0186] in, Represents flexible resources The corresponding utility value, is the total energy cost during period t, is the total load during period t.

[0187] Step 4: Based on the user profile corresponding to the flexible resource, a weighted graph is constructed for each flexible resource;

[0188] The weighted graph uses different types of user portraits corresponding to flexible resources as nodes, and introduces auxiliary nodes; constructs edges based on whether type conversion can be performed between nodes, and the weight of the edge represents the utility change caused by the conversion between flexible resource nodes;

[0189] In a specific implementation, the weighted graph is constructed as follows: different types of user portraits corresponding to flexible resources are used as nodes, and auxiliary nodes are introduced; directed edges are set from auxiliary nodes to other user portrait nodes, and whether directed edges are set between user portrait nodes is determined based on whether type conversion can be performed; each directed edge is set with a weight for indicating the length of the edge, and the weight is determined based on the utility values ​​corresponding to the two nodes.

[0190] Among them, the determination of the weight is specifically as follows: for any first node and second node, the weight of the directed edge from the first node to the second node is the utility value of the second node minus the utility value of the first node; the weight of the directed edge from the second node to the first node is the utility value of the first node minus the utility value of the second node.

[0191] The following is a detailed description of the construction of the weighted graph with specific examples:

[0192] For each flexible resource , construct a weighted graph , the node set is ,in, Flexible resources The type domain (i.e. the category of user portrait), is an auxiliary node, which serves as a unified reference node to make the utility comparison comparable. The set of edges is , the edge weight is defined as follows:

[0193] For the utility value of the auxiliary node, take the lower bound of the utility of all possible user portrait nodes in the weighted graph:

[0194] (28)

[0195] in, Create a user portrait node The corresponding utility value.

[0196] For the edge weight from the auxiliary node to the user portrait node:

[0197] (29)

[0198] For the edge weights between user portrait nodes:

[0199] (30)

[0200] The above formula is expressed as: User portrait node arrive The edge weight is equal to the utility difference between the target node and the starting node.

[0201] Taking electric vehicle load as an example, its type domain includes five typical charging modes: (Fully responsive, flexible charging throughout the day), (Night-time priority, charging during off-peak hours from 23:00 to 7:00), (Working hours type, fixed charging from 9:00 to 17:00), (Emergency protection type, charging is required when the power level is below the threshold), (Fixed mode, charging according to preset time period). In the construction of Indicates from auxiliary node to type The initial utility evaluation reflects the basic value of the user's choice of this type; Indicates that the user is from type Convert to The marginal utility change when For other user types, the type Based on these definitions, a system consisting of six nodes (five types of nodes and one auxiliary node) can be constructed. ), the result is as follows Figure 2 shown.

[0202] Specifically, taking electric vehicle load as an example, Figure 2 Nodes in arrive Representing five charging types of electric vehicle users (corresponding to five user profile types), The blue solid arrow, i.e., the positive weight edge, indicates the increase in utility of the current flexible resource when it is converted from one type to another, i.e., the edge weight value defined previously. ; The red dashed arrow is the negative weight edge, indicating that the utility decreases; the black dotted arrow is the zero weight edge, indicating that the type conversion does not affect the utility.

[0203] Step 5: Based on the weighted graph of each flexible resource, determine the feasible mode for each flexible resource to participate in demand response by calculating the shortest path from the auxiliary node to any user portrait node; based on the feasible modes corresponding to all flexible resources, construct a solution set that meets the current power system needs;

[0204] In the specific implementation, the shortest path from the auxiliary node to any user portrait node is actually expressed as the maximum utility path from the auxiliary node to any portrait node. This is because according to the previous definition, the weight of the edge represents the utility change brought about by the conversion between user portrait nodes, and our goal is to find the path with the maximum utility gain. In order to use the mature shortest path algorithm to solve this maximization problem, the weights of all edges are negative during the calculation, thereby converting the search for the maximum utility path into solving the shortest path problem.

[0205] Specifically, the shortest path is calculated using the Dijkstra algorithm, wherein the Dijkstra algorithm process is as follows: Figure 3 For each flexible resource i, the weights of the directed edges are uniformly negative, and then the Dijkstra algorithm is used to calculate the weights from the auxiliary nodes. To each user portrait node The shortest path is as follows:

[0206] (31)

[0207] This path corresponds to the transformation sequence with the largest utility gain in the original problem, thus determining the optimal response mode of flexible resources.

[0208] Step 6: Based on the obtained solution set, the optimal solution participating in the power system optimization is obtained from the solution set by maximizing the power system optimization objective function.

[0209] Specifically, the above steps 2 to 5 of the present invention are essentially a BOS mechanism (Budget-Optimal Shortest Path Mechanism). The above BOS mechanism design method achieves budget optimization by constructing a weighted graph and calculating the shortest path, while ensuring social efficiency, incentive compatibility and individual rationality, while maintaining stable operation of the power grid and more effectively reducing the implementation cost of demand response.

[0210] The comparison with the existing demand response mechanism design method VCG method is shown in Table 3.

[0211] Table 3 Comparison between BOS mechanism and VCG mechanism

[0212]

[0213] The BOS mechanism achieves social efficiency (SE), dominant strategy incentive compatibility (DSIC) and individual rationality (IR) while minimizing the required budget by cleverly constructing a weighted graph and calculating the shortest path. The BOS mechanism is suitable for power demand response scenarios for the following reasons:

[0214] First, the discreteness and knownness of user types in the power market are highly consistent with the design premise of the BOS mechanism, which can make full use of this information to optimize the mechanism. The BOS mechanism converts the user type domain into vertices in the graph, effectively capturing the discrete characteristics of user preferences.

[0215] Secondly, grid operators need to achieve the best demand response effect within a limited budget, and the budget minimization feature of the BOS mechanism just meets this requirement. Compared with the VCG mechanism, the BOS mechanism can achieve lower budget expenditure in most cases.

[0216] Furthermore, the BOS mechanism cleverly transforms the complex mechanism design problem into a solvable shortest path problem through graph theory methods, which has significant advantages in computational efficiency and scalability. In particular, when the number of users, the number of options, or the size of the type domain increases, the BOS mechanism exhibits good performance, which is conducive to practical application in large-scale power systems.

[0217] Finally, the BOS mechanism not only ensures the maximization of social efficiency, but also realizes the incentive compatibility of the dominant strategy and individual rationality, which is highly consistent with the multiple goals pursued by power demand response. It can effectively balance social interests and individual interests and promote the efficient operation of the power grid.

[0218] Furthermore, the solution described in this embodiment is based on the BOS mechanism described in steps 2 to 5 to calculate the optimal allocation solution. .

[0219] In one or more implementations, corresponding to the above method, this embodiment provides a power system optimization system based on multi-load demand response incentive, including:

[0220] A user portrait generation unit, which is used to perform scenario analysis on the flexible resources on the user side of the power system and generate a user portrait corresponding to the flexible resources; wherein the flexible resources include distributed power sources and flexible loads;

[0221] A utility function construction unit, which is used to construct a corresponding utility function for each user profile of the flexible resource;

[0222] An objective function construction unit, which is used to construct an objective function for optimizing the power system based on the constructed utility function with the goal of maximizing the difference between the utility of all flexible resources and the total energy cost;

[0223] A weighted graph construction unit, which is used to construct a weighted graph for each flexible resource based on the user profile corresponding to the flexible resource;

[0224] A solution set construction unit is used to determine the feasible mode for each flexible resource to participate in demand response based on the weighted graph of each flexible resource by calculating the shortest path from the auxiliary node to any user portrait node; based on the feasible modes corresponding to all flexible resources, a solution set that meets the current power system demand is constructed;

[0225] The optimization solving unit is used to obtain the optimal solution for the power system optimization from the solution set by maximizing the power system optimization objective function based on the obtained solution set.

[0226] It can be understood that the system described in this embodiment corresponds one-to-one to the above method embodiment, and its technical details have been described in detail in the above embodiment, so they will not be repeated here.

[0227] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A power system optimization method based on multi-load demand response incentives, characterized in that: include: Perform scenario analysis on the flexible resources on the user side of the power system to generate user portraits corresponding to the flexible resources; wherein the flexible resources include distributed power sources and flexible loads; For each user profile of the flexible resource, construct the corresponding utility function; Based on the constructed utility function, the power system optimization objective function is constructed with the goal of maximizing the difference between the utility of all flexible resources and the total energy cost; Based on the user portraits corresponding to the flexible resources, a weighted graph is constructed for each flexible resource; The construction of the weighted graph is specifically as follows: different types of user portraits corresponding to flexible resources are used as nodes, and auxiliary nodes are introduced; directed edges are set from auxiliary nodes to other user portrait nodes, and whether directed edges are set between user portrait nodes is determined based on whether type conversion can be performed; each directed edge is set with a weight for indicating the length of the edge, and the weight is determined based on the utility values ​​corresponding to the two nodes; Based on the weighted graph of each flexible resource, the feasible mode for each flexible resource to participate in demand response is determined by calculating the shortest path from the auxiliary node to any user portrait node; The weighted graph introduces auxiliary nodes, which serve as unified reference nodes. For the utility value of the auxiliary node, the lower bound of the utility of all possible user portrait nodes in the weighted graph is taken: ;in, Create a user portrait node The corresponding utility value; for the edge weight from the auxiliary node to the user portrait node: ; For the edge weights between user portrait nodes: ; The above formula is expressed as: User portrait node arrive The edge weight is equal to the utility difference between the target node and the starting node; It is the user portrait type domain of flexible resource i, * is an auxiliary node; Based on the feasible modes corresponding to all flexible resources, a solution set that meets the current power system needs is constructed; Based on the obtained solution set, the optimal solution participating in the power system optimization is obtained from the solution set by maximizing the power system optimization objective function.

2. The power system optimization method based on multi-load demand response incentive according to claim 1, characterized in that: The generation of the user portrait is specifically as follows: for distributed power sources, based on the historical grid-connected power data of distributed power sources, a pre-trained grid-connected power scenario generation model based on deep learning is used to obtain several distributed power source grid-connected scenarios; for flexible loads, based on the historical demand response rate of flexible loads, a Monte Carlo simulation is used to obtain several flexible load response rate scenarios; for the obtained several distributed power source grid-connected scenarios and several flexible load response rate scenarios, a pre-trained variational encoder is used to reduce the scenarios, and the reduced scenarios are used as user portraits of distributed power sources and flexible loads.

3. The power system optimization method based on multi-load demand response incentive according to claim 1, characterized in that: The utility function is used to quantify the utility values ​​corresponding to different user profiles of flexible resources, and to quantify the utility values ​​into price indicators by introducing standard electricity prices, wherein utility parameters are introduced into the utility function, and the utility parameters are determined based on the priority coefficient of the flexible resources participating in demand response, the standard electricity price, and the product of the response characteristic function.

4. The power system optimization method based on multi-load demand response incentive according to claim 1, characterized in that: The utility function is specifically expressed as follows: in, Represents flexible resources The corresponding utility value; Represents flexible resources The amount of electricity or electricity demand in the grid during period t, is the utility parameter, which is expressed as the product of the priority coefficient of the flexible resource participating in the demand response, the standard electricity price and the response characteristic function; To adjust the parameters.

5. The power system optimization method based on multi-load demand response incentive according to claim 1, characterized in that: The shortest path from the auxiliary node to any user portrait node aims to solve the maximum utility path from the auxiliary node to any portrait node, and the shortest path is solved using the Dijkstra algorithm.

6. The power system optimization method based on multi-load demand response incentive according to claim 1, characterized in that: The power system optimization objective function is specifically expressed as follows: in, Represents flexible resources The corresponding utility value, is the total energy cost during period t, is the total load during period t.

7. The power system optimization system based on multi-load demand response incentive is characterized by: include: A user portrait generation unit, which is used to perform scenario analysis on the flexible resources on the user side of the power system and generate a user portrait corresponding to the flexible resources; wherein the flexible resources include distributed power sources and flexible loads; A utility function construction unit, which is used to construct a corresponding utility function for each user profile of the flexible resource; An objective function construction unit, which is used to construct an objective function for optimizing the power system based on the constructed utility function with the goal of maximizing the difference between the utility of all flexible resources and the total energy cost; A weighted graph construction unit, which is used to construct a weighted graph for each flexible resource based on the user profile corresponding to the flexible resource; The construction of the weighted graph is specifically as follows: different types of user portraits corresponding to flexible resources are used as nodes, and auxiliary nodes are introduced; directed edges are set from auxiliary nodes to other user portrait nodes, and whether directed edges are set between user portrait nodes is determined based on whether type conversion can be performed; each directed edge is set with a weight for indicating the length of the edge, and the weight is determined based on the utility values ​​corresponding to the two nodes; The solution set construction unit is used to determine the feasible mode for each flexible resource to participate in demand response based on the weighted graph of each flexible resource by calculating the shortest path from the auxiliary node to any user portrait node; based on the feasible modes corresponding to all flexible resources, a solution set that meets the current power system demand is constructed; the weighted graph introduces auxiliary nodes, which serve as unified reference nodes. For the utility value of the auxiliary nodes, the lower bound of the utility of all possible user portrait nodes in the weighted graph is taken: ;in, Create a user portrait node The corresponding utility value; for the edge weight from the auxiliary node to the user portrait node: ; For the edge weights between user portrait nodes: ; The above formula is expressed as: User portrait node arrive The edge weight is equal to the utility difference between the target node and the starting node; It is the user portrait type domain of flexible resource i, * is an auxiliary node; The optimization solving unit is used to obtain the optimal solution for the power system optimization from the solution set by maximizing the power system optimization objective function based on the obtained solution set.

8. The power system optimization system based on multi-load demand response incentive according to claim 7, characterized in that: The generation of the user portrait is specifically as follows: for distributed power sources, based on the historical grid-connected power data of distributed power sources, a pre-trained grid-connected power scenario generation model based on deep learning is used to obtain several distributed power source grid-connected scenarios; for flexible loads, based on the historical demand response rate of flexible loads, a Monte Carlo simulation is used to obtain several flexible load response rate scenarios; for the obtained several distributed power source grid-connected scenarios and several flexible load response rate scenarios, a pre-trained variational encoder is used to reduce the scenarios, and the reduced scenarios are used as user portraits of distributed power sources and flexible loads.

9. The power system optimization system based on multi-load demand response incentive according to claim 7, characterized in that: The utility function is used to quantify the utility values ​​corresponding to different user profiles of flexible resources, and to quantify the utility values ​​into price indicators by introducing standard electricity prices, wherein utility parameters are introduced into the utility function, and the utility parameters are determined based on the priority coefficient of the flexible resources participating in demand response, the standard electricity price, and the product of the response characteristic function.

10. The power system optimization system based on multi-load demand response incentive according to claim 7, characterized in that: The utility function is specifically expressed as follows: in, Represents flexible resources The corresponding utility value; Represents flexible resources The amount of electricity or electricity demand in the grid during period t, is the utility parameter, which is expressed as the product of the priority coefficient of the flexible resource participating in the demand response, the standard electricity price and the response characteristic function; To adjust the parameters.

Images