Meteorological data driven demand side resource real-time rolling response optimization method

Abstract

The invention relates to a meteorological data-driven demand side resource real-time rolling response optimization method, which belongs to the field of comprehensive energy, and comprises the following steps of: preprocessing acquired meteorological data in a certain historical period of an area to be optimized and corresponding user behavior data to obtain a training sample set; training by using the training sample set to obtain a trained correlation analysis model; inputting the acquired meteorological data of the adjustment time period of the operation day into the trained correlation analysis model to obtain predicted demand side user behavior data of the adjustment time period; obtaining a corresponding demand side user load based on the predicted demand side user behavior data of the adjustment time period, and constructing a demand side resource response capability model based on the demand side user load; carrying out real-time rolling solving on the demand side resource response capability model to obtain a response plan of the demand side resources in the adjustment time period; and performing real-time rolling solution on each adjustment time period in the adjustment period in sequence to obtain a demand side resource response plan of the demand side resources in the region in the adjustment period.

Description

Technical Field

[0001] The present invention relates to the field of integrated energy technology, and in particular to a meteorological data-driven demand-side resource real-time rolling response optimization method. Background Art

[0002] A high proportion of renewable energy integration will become a fundamental feature of future power systems. The combined challenges of integrating renewable energy and ensuring power supply will lead to a sharp increase in the demand for regulation in the power system. Therefore, tapping into a larger pool of flexible resources has become a primary challenge in building future power systems. Demand-side resources such as electric vehicles and air conditioning loads offer advantages such as high flexibility and rapid response. They are widely distributed across different regions and usage scenarios within the power system, and their flexibility and regulation potential offer new solutions to challenges such as integrating renewable energy and ensuring power supply.

[0003] Research on demand-side resource response optimization methods primarily focuses on aggregating and utilizing various demand-side resources through methods such as virtual power plants. These methods, based on the system's regulation needs, determine the response output plans for each demand-side resource based on the expected response capabilities of each resource, with the goal of maximizing the satisfaction of the power system's regulation needs. However, during real-time operation, meteorological data such as temperature and precipitation can cause certain deviations in the response capabilities of demand-side resources such as electric vehicles and air conditioning loads. Existing research, however, has not addressed the impact of meteorological data and user behavior data on the real-time response capabilities of demand-side resources. This makes it impossible to formulate real-time response plans for each demand-side resource based on the deviations in the real-time response capabilities of the aggregated response entities, placing significant pressure on the power system. Summary of the Invention

[0004] In view of the above analysis, an embodiment of the present invention aims to provide a meteorological data-driven demand-side resource real-time rolling response optimization method to solve the technical problem of demand resource response deviation caused by meteorological data and user behavior data in the existing technology.

[0005] The present invention provides a method for optimizing the real-time rolling response of demand-side resources driven by meteorological data, comprising the following steps:

[0006] Preprocessing the acquired meteorological data and corresponding user behavior data within a certain historical period of the area to be optimized to obtain a training sample set, and using the training sample set to train a trained association analysis model; inputting the acquired meteorological data during the adjustment period of the operating day into the trained association analysis model to obtain predicted demand-side user behavior data during the adjustment period;

[0007] Obtaining corresponding demand-side user load based on the predicted demand-side user behavior data during the adjustment period, and building a demand-side resource responsiveness model based on the demand-side user load;

[0008] The demand-side resource response capability model is solved in real time and rollingly to obtain the response plan of the demand-side resources in the adjustment period; and each adjustment period in the adjustment cycle is solved in real time and rollingly in turn to obtain the demand-side resource response plan of the demand-side resources in the region within the adjustment period.

[0009] Furthermore, the pre-processing of the acquired meteorological data within a certain historical period and the corresponding user behavior data in the area to be optimized to obtain a training sample set includes:

[0010] Select user behavior data determined by meteorological data;

[0011] Based on the selected user behavior data and corresponding meteorological data, data within an invalid period is eliminated to obtain meteorological data and corresponding user behavior data within a valid period as the first meteorological data and first user behavior data; wherein the valid period is a time period within the scheduling period during which user behavior data is sensitive to changes in meteorological data and has an actual response;

[0012] Calculating the Pearson correlation coefficient between the first meteorological data and the first user behavior data, and selecting strongly correlated meteorological data and user behavior data with a correlation coefficient greater than or equal to a preset correlation threshold as the second meteorological data and the second user behavior data;

[0013] Calculating the correlation between the operating day meteorological data and the second meteorological data using a grey correlation analysis method, and selecting meteorological data and corresponding user behavior data with a correlation greater than or equal to a preset correlation threshold as the third meteorological data and third user behavior data;

[0014] Calculating a ratio of a distance from the third user behavior data to the sample center to an average distance, and removing third user behavior data having the ratio greater than a preset abnormality threshold as an outlier, thereby obtaining fourth meteorological data and fourth user behavior data from which the outliers have been removed, to form the training sample set;

[0015] The fourth user behavior data includes continuity and time-based user behavior data.

[0016] Further, the Pearson correlation coefficient between the first meteorological data and the first user behavior data is calculated.

[0017] Select The first meteorological data and the corresponding first user behavior data that are greater than or equal to the preset correlation threshold are used as the second meteorological data and the second user behavior data.

[0018] Furthermore, the association analysis model includes a multiple linear regression sub-model and a logistic regression sub-model;

[0019] The fourth meteorological data is used as the independent variable and the fourth continuous user behavior data is used as the dependent variable to establish a multivariate linear regression sub-model;

[0020] The fourth meteorological data is used as the independent variable and the fourth user behavior data corresponding to the time type is used as the dependent variable to establish a logistic regression sub-model.

[0021] Furthermore, the association analysis model is trained using the training sample set to obtain a trained association analysis model, including:

[0022] The least squares method was used to fit the coefficients in the multivariate linear regression sub-model, and the fitted coefficients were obtained as follows:

[0023]

[0024] in, is the coefficient of the fitted multiple linear regression sub-model;

[0025] The cross entropy loss function is used as the cost function of the logistic regression sub-model to measure the data fitting degree of the logistic regression function. The loss function is as follows:

[0026]

[0027] Among them, J(τ) is the loss function; τ is the learning rate, Y f and Y r are the predicted value and actual value of user behavior data respectively; n H2 is the sample size of the time-based fourth user behavior data and the corresponding fourth meteorological data;

[0028] The training sample set is used to iteratively optimize a and τ using the gradient descent method until the joint loss function of the multivariate linear regression sub-model and the logistic regression sub-model converges, thereby obtaining a trained association analysis model.

[0029] Furthermore, the expected load power of each demand-side resource must be within the elastic operating boundary of the demand-side resource on the operation day, as follows:

[0030]

[0031] in, is the expected load power of demand-side resource i in regulation period t; P i is the elastic operation boundary of demand-side resource i from the current moment to the end of the adjustment period;

[0032] The elastic operating boundary of the demand-side resources on the operating day is determined by the following steps:

[0033] Based on the latest operational day meteorological data at the current moment, the trained association analysis model is input to predict the demand-side user behavior data as follows;

[0034]

[0035] in, is the j-th user behavior data of demand-side resource i; Ψ(·) is the process of predicting user behavior data based on the latest operating day meteorological data by the association analysis model; MI k is the kth meteorological data corresponding to the user behavior data; is the meteorological data set that affects the behavior data of the jth user of demand-side resource i; Ω i is the user set of resource i on the demand side;

[0036] Based on the demand-side resource response capability quantification model and the predicted demand-side user behavior data, the upper and lower limits of the operating power of each demand-side resource within the adjustment period are calculated;

[0037]

[0038] where Φ(·) represents the constraint imposed by the quantitative model of the demand-side resource response capability on the load power of the demand-side resources in each period from the current moment to the end of the regulation period.

[0039] The calculated upper and lower limits of the operating power within the adjustment period are used as elastic operating boundaries and sent synchronously to the demand-side user to constrain the real-time response plan of the demand-side resources.

[0040] Furthermore, the real-time rolling response optimization solution of the demand-side resource response capability model includes:

[0041] With the goal of maximizing the power system regulation demand, the objective function and constraints of the real-time rolling response optimization solution are set;

[0042] The planned load power of each demand-side resource in the regulation period t is used as the initial value of the demand-side response plan, as follows:

[0043]

[0044] Among them, P t r is the vector of each demand-side resource response plan in the regulation period t, n dr is the number of demand-side resources;

[0045] The demand-side resource response capability model is solved using the CPLEX solver. The adjustment period in the adjustment cycle of the operation day is t ad , from the adjustment period t ad-n, based on the latest operational day meteorological data, the response plan of each demand-side resource is dynamically adjusted every q hours; one adjustment cycle includes multiple adjustment periods;

[0046] During the rolling optimization process, the expected load power of each demand-side resource is within the elastic operating boundary, as follows:

[0047]

[0048] in, is the expected load power of demand-side resource i in regulation period t;

[0049] The model is solved every q hours to dynamically update the resource response plan for each demand side;

[0050] The response plan of each demand-side resource during the adjustment period is updated and sent synchronously to the demand-side user as follows:

[0051]

[0052] Until the adjustment period t ad The demand-side resource response plan has actually occurred, and the demand-side resources are adjusted during the adjustment period t ad The rolling optimization of the response plan has been completed;

[0053] In the first n hours of the next regulation period, the rolling optimization steps are repeated until the demand-side resource response plans for all regulation periods within the scheduling cycle have actually occurred, and the demand-side resource response behavior of the demand-side resources in the region in each regulation period within the regulation cycle is obtained.

[0054] Furthermore, the objective function of the real-time rolling response optimization solution is as follows:

[0055]

[0056] in, To meet the demand response capability of the power system in the region; R ad,t and F pun,t are the settlement income and deviation penalty fees in adjustment period t respectively; and C loss,i,t are the lth economic response cost and electricity satisfaction loss cost of demand-side resource i in regulation period t; I is the demand-side resource set.

[0057] Furthermore, the constraints of the real-time rolling response optimization solution include the elastic operating boundaries of demand-side resources, as well as electric vehicle, air-conditioning load, industrial load and user-side energy storage constraints.

[0058] Furthermore, the demand-side user load includes electric vehicles, air-conditioning loads, industrial loads and user-side energy storage loads;

[0059] The demand-side resource response capability model includes electric vehicle, air conditioning load, industrial load and user-side energy storage response capability models.

[0060] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0061] 1. By introducing a correlation analysis model based on meteorological data and user behavior data, this method can more accurately predict the responsiveness of demand-side resources. Compared with traditional methods, this method can adjust the response plan of demand-side resources in real time, reduce the deviation of demand-side resource response caused by changes in meteorological data, and improve the regulation accuracy of the power system based on accurate prediction driven by meteorological data.

[0062] 2. This invention dynamically adjusts the response plan of each demand-side resource within the regulation cycle by solving the demand-side resource response capability model in real time. This rolling optimization mechanism can timely update the demand-side resource response plan based on the latest meteorological data and user behavior data to ensure that the regulation needs of the power system are maximized.

[0063] 3. This invention introduces the concept of flexible operating boundaries for demand-side resources, ensuring that the expected load power of each demand-side resource fluctuates within a reasonable range. This constraint mechanism not only ensures the stability of the power system, but also improves the utilization efficiency of demand-side resources and avoids over-regulation or under-regulation.

[0064] 4. This invention comprehensively considers the responsiveness of multiple demand-side resources, including electric vehicles, air conditioning loads, industrial loads, and user-side energy storage, to construct a multi-type demand-side resource responsiveness model. By collaboratively optimizing these multiple demand-side resources, the system leverages flexibility and enhances the overall regulation capability of the power system.

[0065] In the present invention, the above-mentioned technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of the present invention will be described in the following description, and some advantages will become apparent from the description or be learned through practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the contents particularly pointed out in the description and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] The accompanying drawings are only for the purpose of illustrating particular embodiments and are not to be considered limiting of the present invention. Like reference symbols denote like parts throughout the drawings.

[0067] Figure 1This is a flow chart of a meteorological data-driven demand-side resource real-time rolling response optimization method according to an embodiment of the present invention;

[0068] Figure 2 Schematic diagram of real-time response of demand-side resources based on meteorological data during the adjustment period of an operating day in an embodiment of the present invention;

[0069] Figure 3 This is a flowchart of the correlation analysis between meteorological data and user behavior data in an embodiment of the present invention;

[0070] Figure 4 This is a flowchart for rolling update of the response plan of demand-side resources in an embodiment of the present invention. DETAILED DESCRIPTION

[0071] The preferred embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, and are not used to limit the scope of the present invention.

[0072] The present invention clarifies the impact of meteorological data and user behavior data on the real-time response capability of demand-side resources, focuses on the real-time response capability deviation of each demand-side resource in the real-time response stage, studies the real-time rolling optimization model of demand-side resources, and dynamically adjusts the real-time response plan of each demand-side resource based on the latest meteorological data and user behavior data to ensure that demand-side resources can fully respond to demand-side resource needs and avoid waste of power resources.

[0073] The present invention constructs a correlation analysis model between meteorological data and user behavior data and a real-time rolling optimization model for demand-side resources. In actual operation, based on the meteorological data of each adjustment period on the latest operating day, the real-time response plan of each resource is continuously updated and adjusted. This can effectively ensure that demand-side resources receive a timely response, avoid unnecessary waste of power resources, improve energy utilization, and provide direction and basis for future research on demand-side resource response optimization methods.

[0074] A specific embodiment of the present invention, as Figure 1 As shown, a method for optimizing the real-time rolling response of demand-side resources driven by meteorological data is disclosed, comprising the following steps:

[0075] Step S1: Preprocess the acquired meteorological data and corresponding user behavior data of the area to be optimized within a certain historical period to obtain a training sample set, and use the training sample set to train a trained association analysis model; input the acquired meteorological data of the adjustment period of the operating day into the trained association analysis model to obtain the predicted demand-side user behavior data of the adjustment period;

[0076] Step S2: obtaining corresponding demand-side user load based on the predicted demand-side user behavior data during the adjustment period, and constructing a demand-side resource responsiveness model based on the demand-side user load;

[0077] Step S3: Perform real-time rolling solution on the demand-side resource response capability model to obtain the demand-side resource response plan for the adjustment period; and perform real-time rolling solution on each adjustment period in the adjustment cycle in turn to obtain the demand-side resource response plan for the demand-side resources in the region within the adjustment cycle.

[0078] The step S1 includes steps S11-S12.

[0079] Step S11, pre-processing the acquired meteorological data within a certain historical period and the corresponding user behavior data in the area to be optimized to obtain a training sample set, includes:

[0080] Select user behavior data determined by meteorological data;

[0081] Based on the selected user behavior data and corresponding meteorological data, data within an invalid period is eliminated to obtain meteorological data and corresponding user behavior data within a valid period as the first meteorological data and first user behavior data; wherein the valid period is a time period within the scheduling period during which user behavior data is sensitive to changes in meteorological data and has an actual response;

[0082] Calculating the Pearson correlation coefficient between the first meteorological data and the first user behavior data, and selecting strongly correlated meteorological data and user behavior data with a correlation coefficient greater than or equal to a preset correlation threshold as the second meteorological data and the second user behavior data;

[0083] Calculating the correlation between the operating day meteorological data and the second meteorological data using a grey correlation analysis method, and selecting meteorological data and corresponding user behavior data with a correlation greater than or equal to a preset correlation threshold as the third meteorological data and third user behavior data;

[0084] Calculating a ratio of a distance from the third user behavior data to the sample center to an average distance, and removing third user behavior data having the ratio greater than a preset abnormality threshold as an outlier, thereby obtaining fourth meteorological data and fourth user behavior data from which the outliers have been removed, to form the training sample set;

[0085] The fourth user behavior data includes continuity and time-based user behavior data.

[0086] Meteorological data such as temperature and precipitation affect user behavior data, which in turn affects the response capability of the power system. This paper analyzes the correlation between a large number of meteorological data and user behavior data of various resources determined by objective factors, and uses multiple linear regression and logistic regression methods to establish correlation analysis models between meteorological data and continuous user behavior data and time-based user behavior data, such as Figure 3 shown.

[0087] (1) Select user behavior data determined by meteorological data.

[0088] Examples of user behavior data for electric vehicles, air-conditioning loads, industrial loads, and user-side energy storage are shown in Table 1. Items not marked with an * represent user behavior data determined by objective factors of meteorological data, which are often affected by meteorological data such as temperature and rainfall. This user behavior data can be added or deleted based on specific actual needs. Other items marked with an * represent behavioral parameters that can be subjectively controlled by the user. Considering that the user behavior data of industrial loads and user-side energy storage are relatively fixed, and since industrial loads and user-side energy storage are basically unaffected by meteorological data, this invention mainly studies the correlation between meteorological data and user behavior data of electric vehicles and air-conditioning loads.

[0089] Table 1: User behavior data for various demand-side resources

[0090]

[0091]

[0092] (2) Select meteorological data corresponding to user behavior data.

[0093] Exemplarily, the present invention is applied to industrial areas, and user behavior data comes from historical operating data recorded by users' smart meters, smart energy management systems and other equipment; meteorological data comes from predicted real-time meteorological data released by the China Meteorological Network. The various types of meteorological data selected are specifically shown in Table 2.

[0094] Table 2: Description of various meteorological data

[0095]

[0096] (3) Eliminate the data in the invalid period and obtain the meteorological data and the corresponding user behavior data in the valid period as the first meteorological data and the first user behavior data.

[0097] For user behavior data, some adjustment periods within a scheduling cycle (a scheduling cycle includes one or more adjustment periods) are often invalid, and data from these invalid periods is discarded. For example, electric vehicle users often do not travel in the early morning, so weather data during these adjustment periods generally does not affect user behavior data.

[0098] The meteorological data in Table 2 is calculated only from the data during the valid period of user behavior data, as follows:

[0099]

[0100] in, A max and A min are the average value, highest value (or maximum value) and lowest value (or minimum value) of a certain meteorological data in the valid period of a certain historical date; A t is the value of the meteorological data during the valid period of the historical date; Z is the valid period set of the user behavior data; n Z is the number of valid periods of meteorological data; n is the number of meteorological data k, which has n types; A k (t), are the meteorological data and the average value of the meteorological data in period t respectively;

[0101] To eliminate the impact of user behavior data fluctuations and seasonal weather data, historical sample data is collected from a certain period before the launch date. Furthermore, to distinguish the impact of date type on user behavior data, the input data for the association analysis model is data from the historical sample data (including weather data and user behavior data) that has the same date type as the launch date.

[0102] Historical data is divided into two main date types: 1) weekdays; 2) weekends and statutory holidays. User behavior data and operating day data have the same date type. For example, if the operating day is Monday, user behavior data will be historical data from the previous weekday from Monday to Friday.

[0103] (4) Calculate the Pearson correlation coefficient between the first meteorological data and the first user behavior data, and select strongly correlated meteorological data and user behavior data whose correlation coefficient is greater than or equal to a preset correlation threshold as the second meteorological data and the second user behavior data.

[0104] Meteorological data is diverse and complex, and the correlation characteristics of various types of meteorological data differ from those of demand-side resource user behavior data. Incorporating this data into correlation analysis models for training not only increases computational complexity but also reduces model reliability. Before performing correlation analysis, irrelevant meteorological data features must be eliminated.

[0105] The Pearson correlation coefficient (Pearson product-moment correlation coefficient) is used to describe the correlation between meteorological data and demand-side resource user behavior data. The historical meteorological data set is X d (k), k=1,2,…,nd∈D, n is the meteorological data type; D is the historical date set; the historical data set of a user behavior data is Y d ,d∈D; including the historical data of the user's electric vehicle, air conditioning load, industrial load and user-side energy storage.

[0106] For example, considering the weather changes caused by seasonal changes, data from one month ago is selected.

[0107] Calculate the Pearson correlation coefficient between the first meteorological data and the first user behavior data

[0108] Select The first meteorological data and the corresponding first user behavior data that are greater than or equal to the preset correlation threshold are used as the second meteorological data and the second user behavior data. Pearson correlation coefficient as follows:

[0109]

[0110] Where cov(X(k), Y) is the covariance of X(k) and Y; X(k) and Y are the first meteorological data and the first user behavior data sample in the training sample set respectively; σ Y X d (k) and Y d The standard deviation of X d (k) is the first meteorological data set; Y d is the corresponding first user behavior dataset; D is the set of dates in the selected historical period; and X d (k) and Y d The mean value of , n is the meteorological data type;

[0111] The value range of is [-1,1], A larger absolute value indicates a higher correlation.

[0112] when When the value of is between [0,0.2], it means that X(k) and Y are unrelated;

[0113] when When the value of is between [0.2, 0.4], it means that X(k) and Y are weakly correlated;

[0114] when When the value of is between [0.4, 0.6], it means that X(k) and Y are moderately correlated;

[0115] when When the value of is between [0.6, 0.8], it means that X(k) and Y are strongly correlated;

[0116] when When the value of is between [0.8,1], it means that X(k) and Y are strongly correlated.

[0117] Repeatedly calculate the correlation coefficient between each meteorological data and user behavior data. Select meteorological data with strong correlation or above to construct a typical meteorological data set that affects the user behavior data, denoted as K.

[0118] For example, the preset correlation threshold is 0.6; Weather data; exclusion of meteorological data.

[0119] For example, assume that there are five types of meteorological data: temperature, wind speed, precipitation, air pressure, and cloud cover. Among them, temperature and wind speed are strongly correlated with user behavior data of electric vehicle charging time, and the other three are all less than 0.6. Then the meteorological data set that affects the charging time of electric vehicles is K = {temperature, wind speed}.

[0120] Filter out meteorological data related to demand-side resource user behavior data from a large amount of meteorological data; and use the selected meteorological data and corresponding user behavior data as second meteorological data and second user behavior data.

[0121] (5) Using the grey correlation analysis method to calculate the correlation between the operating day meteorological data and the second meteorological data, the meteorological data and the corresponding user behavior data with a correlation greater than or equal to a preset correlation threshold are selected as the third meteorological data and the third user behavior data.

[0122] Both meteorological data and user behavior data are multidimensional and discrete. The impact of meteorological data on user behavior data is complex and difficult to quantify directly. However, under the same weather conditions, user behavior data tends to be relatively similar, making it more suitable for correlation analysis.

[0123] Therefore, the similarity between the meteorological data of the operation day and the historical meteorological data is calculated. If the meteorological data of some nearby dates in the historical data are close to that of the operation day, the data of these dates are considered to be similar to that of the operation day.

[0124] The correlation degree is used to measure the matching degree between two objects (the meteorological data of the operation day and the meteorological data of the historical date). The grey correlation analysis method is used to calculate the correlation degree between the meteorological data of the operation day and the meteorological data of the historical date, and the historical date data with the higher correlation degree is selected.

[0125] The calculation process of the grey relational analysis method is as follows:

[0126] The first step is to record the meteorological data of the operation day as X0 = {x0(k) | k = 1, 2, ..., K}; record the second meteorological data of the historical date as X d ={x d (k)|k=1,2,…,K},d∈D.

[0127] Calculate the correlation coefficient ξ between the kth meteorological data of the operation day and the second meteorological data of each historical date d (k) as follows:

[0128]

[0129] Where γ is the resolution coefficient. The smaller γ is, the greater the resolution is. For example, the resolution is best when γ≤0.5463, which is 0.5.

[0130] The second step is to calculate the correlation coefficient between a certain meteorological data on the operation day and the second meteorological data on the historical date. d (k) The cumulative average value is used as the correlation between the meteorological data of the operation day and the second meteorological data of the historical date, calculated as follows:

[0131]

[0132] Among them, r GRA,d The operating day meteorological data X0 and the second meteorological data X0 of the historical period d The correlation degree; n K is the number of typical meteorological data of the user behavior data, k , K are the second meteorological data and the second meteorological data set respectively.

[0133] The third step is to repeatedly calculate the correlation between the meteorological data of the operation day and the meteorological data of each historical date, and arrange the correlation between the meteorological data of each historical date in order.

[0134] Set the correlation threshold r GRA,0 , select the meteorological data of the operation day and the meteorological data of the historical date whose correlation is greater than the threshold r GRA,0 The historical date data (historical meteorological data, historical user behavior data) is used as the third meteorological data and the third user behavior data. The correlation degree is less than the threshold r GRA,0Historical date data is excluded.

[0135] The correlation between the meteorological data of the operation day and the meteorological data of each historical period is mostly concentrated between 0.6 and 1.0. For example, the preset correlation threshold r GRA,0 Set to 0.6.

[0136] There is considerable uncertainty in user behavior for electric vehicles and air conditioning loads. Outliers in user behavior data often differ significantly from other parameter values ​​in the training sample set. If not removed, they will significantly impact the accuracy of the association analysis model.

[0137] (6) Calculate the ratio of the distance from the third user behavior data to the sample center to the average distance, and remove the third user behavior data whose ratio is greater than a preset abnormality threshold as an outlier, thereby obtaining the fourth meteorological data and the fourth user behavior data from which the outliers are removed, and forming the training sample set.

[0138] The fourth user behavior data includes continuity and time-based user behavior data.

[0139] Calculate the ratio of the distance from the third user data to the sample center to the average distance o h , identify the abnormal values ​​of the third user's user behavior data as follows:

[0140]

[0141] Among them, n H is the number of training sample sets; if o h If it is greater than the preset abnormal threshold o0, the sample h is regarded as an abnormal value and removed from the training sample set, and o is retained. h The sample data that is less than or equal to the threshold o0 is used to obtain the fourth meteorological data and the fourth user behavior data after the abnormal values ​​are eliminated.

[0142] Exemplarily, the preset outlier threshold o0 is set to 3. If the ratio of the distance from a third user data sample to the sample center to the average distance is greater than 3, the third user data sample is considered as an outlier and is removed.

[0143] The function of step S11 is to construct a training sample set through data screening, cleaning and correlation analysis.

[0144] Step S12: using the training sample set to train and obtain a trained association analysis model; inputting the acquired meteorological data of the adjustment period of the operation day into the trained association analysis model to obtain the predicted demand-side user behavior data of the adjustment period.

[0145] Because multiple linear regression and logistic regression require relatively low data volumes and have high training speeds, they are well-suited to the real-time interaction between demand-side resources for electric vehicles and air conditioning loads. This invention uses multiple linear regression to establish correlation analysis models between meteorological data and continuous user behavior data, such as the remaining battery capacity during grid-connected electric vehicles. Furthermore, multiple logistic regression is used to establish correlation analysis models between meteorological data and temporal user behavior data, such as the periods during which electric vehicles are grid-connected and the periods during which air conditioning loads are activated.

[0146] The association analysis model includes a multiple linear regression sub-model and a logistic regression sub-model;

[0147] The fourth meteorological data is used as the independent variable and the fourth continuous user behavior data is used as the dependent variable to establish a multivariate linear regression sub-model;

[0148] The fourth meteorological data is used as the independent variable and the fourth user behavior data corresponding to the time type is used as the dependent variable to establish a logistic regression sub-model.

[0149] The pre-processed fourth user behavior data is classified into continuous user behavior data and temporal user behavior data. Multiple linear regression and logistic regression are performed on the continuous and temporal fourth user behavior data, as well as the corresponding fourth meteorological data.

[0150] For example, the remaining power of an electric vehicle when connected to the grid, the cooling capacity of the air conditioner load, the indoor and outdoor temperature difference, and the rated cooling power of the air conditioner are all continuous values, which are continuous user behavior data;

[0151] The electric vehicle grid-connected period is the specific time point when the electric vehicle starts to be connected to the grid for charging. It is usually discrete and is a specific hour or time period. The air-conditioning load start-up period is the specific time point when the air-conditioning is turned on. It is a discrete time value. The number of user behaviors occurs within a certain period of time. The number of times a user performs a certain operation, such as starting an electric vehicle or turning on the air-conditioning, is counted data and is regarded as time-type user behavior data. It is associated with the time point and is a discrete value.

[0152] (1) Multiple linear regression sub-model: Taking the fourth meteorological data as the independent variable and the continuous fourth user behavior data as the dependent variable, the expression of the multiple linear regression is as follows:

[0153]

[0154] Among them, Y is the dependent variable, which is the fourth user behavior data; a k is the linear term coefficient of the kth meteorological characteristic variable; X(k) is the independent variable, which is the fourth meteorological data; a0 is the constant term coefficient; ε is the error term.

[0155] (2) Logistic regression sub-model: Logistic regression is suitable for processing binary classification problems. By mapping the results of the logistic regression model to probability values ​​in the range of [0, 1], it can effectively perform classification prediction on discrete user behavior data.

[0156] For the time-type fourth user behavior data, firstly, a regular value set of the fourth user behavior data is constructed, that is, a set consisting of the user behavior data values ​​of all training samples.

[0157] A corresponding logistic regression sub-model is established for each adjustment period, and the probability value of user behavior occurring in each period is calculated respectively. The period with the largest probability value is determined as the prediction result of user behavior occurrence.

[0158] The specific steps of logistic regression are as follows:

[0159] The first step is to use the fourth meteorological data as the independent variable and the time-based fourth user behavior data as the dependent variable to establish a logical function as follows:

[0160]

[0161] Wherein, g(Z) is a logic function, illustratively, the logic function is a sigmoid function; Z is an intermediate variable linearly mapped by various meteorological data.

[0162] The association analysis model is trained using the training sample set to obtain a trained association analysis model, including:

[0163] The least squares method was used to fit the coefficients in the multivariate linear regression sub-model, and the fitted coefficients were obtained as follows:

[0164]

[0165] in, is the coefficient of the fitted multiple linear regression sub-model;

[0166] The cross entropy loss function is used as the cost function of the logistic regression sub-model to measure the data fitting degree of the logistic regression function. The loss function is as follows:

[0167]

[0168] Among them, J(τ) is the loss function; τ is the learning rate, Y f and Y r are the predicted value and actual value of user behavior data respectively; n H2 is the sample size of the time-based fourth user behavior data and the corresponding fourth meteorological data;

[0169] The training sample set is used to iteratively optimize a and τ using the gradient descent method until the joint loss function of the multivariate linear regression sub-model and the logistic regression sub-model converges, thereby obtaining a trained association analysis model.

[0170] The loss function of the multivariate linear regression sub-model uses the mean square error MSE, as follows:

[0171]

[0172] Among them, n H1 is the sample size of the continuous fourth user behavior data and the corresponding fourth meteorological data.

[0173] The joint loss function is adopted as follows:

[0174] L jiont =ω1L linear +ω2J(τ) Formula (13)

[0175] Among them, ω1 and ω2 are the weights of the multivariate linear regression and logistic regression loss functions, respectively.

[0176] Based on the fourth meteorological data and the fourth user behavior data in the training sample set, the gradient descent method is used to iteratively optimize a and τ until the joint function converges to obtain a trained association analysis model.

[0177] The representation of user behavior data in the association analysis model at different time periods has certain differences.

[0178] Taking electric vehicle grid connection behavior as an example, suppose a user's grid connection periods from days 1 to 10 are [6, 7, 6, 8, 5, 6, 8, 6, 7, 6]. For period 5, the user's grid connection behavior is expressed as [0, 0, 0, 0, 1, 0, 0, 0, 0]; for period 6, it is [1, 0, 1, 0, 0, 1, 0, 1]; for period 7, it is [0, 1, 0, 0, 0, 0, 0, 1, 0]; and for period 8, it is [0, 0, 0, 1, 0, 0, 1, 0, 0, 0].

[0179] The meteorological data of the regulation period on the operating day is input into the trained association analysis model to obtain the demand-side user behavior data of the predicted regulation period.

[0180] The purpose of step S12 is to predict the demand-side user behavior data during the adjustment period by training the association analysis model.

[0181] Step S1 is to predict demand-side user behavior data during the adjustment period through data preprocessing, building a training sample set, and training an association analysis model. This provides data support for subsequent demand-side resource response capability modeling and real-time rolling optimization.

[0182] Step S2, specifically.

[0183] The corresponding demand-side user load is obtained based on the predicted demand-side user behavior data of the adjustment period, and a demand-side resource responsiveness model is constructed based on the demand-side user load.

[0184] The demand-side user loads include electric vehicles, air-conditioning loads, industrial loads and user-side energy storage loads;

[0185] The demand-side resource response capability model includes electric vehicle, air conditioning load, industrial load and user-side energy storage response capability models.

[0186] (1) Electric vehicle response capability model.

[0187] Electric vehicle status is divided into five categories: normal charging, fast charging, discharging (reverse charging to the grid, different from traveling), waiting (unused standby state) and traveling. The orderly management of electric vehicle charging and discharging behavior provides regulated output for the grid.

[0188] In the grid-connected state, the remaining power and operating power of the electric vehicle are as follows:

[0189]

[0190] in, are the remaining power of electric vehicle i in time period t and t+1 respectively; is the operating power of electric vehicle i in time period t; and It is the fast charging power, normal charging power and discharge power of electric vehicles. and They are 0-1 variables that control fast charging, normal charging, and discharging states respectively; and They are the fast charging efficiency, normal charging efficiency and discharge efficiency of electric vehicles respectively.

[0191] The above formula represents the change of electric vehicle power in the grid-connected state. In order to control the grid-connected and off-grid states of electric vehicles, a 0-1 variable is introduced. Indicates the grid-connected and off-grid status of electric vehicles, and satisfies the following grid-connected and off-grid constraints of electric vehicles:

[0192]

[0193] in, When it is 1, it means that the electric vehicle is in the grid-connected state. At this time, it can be in fast charging, normal charging, discharging and waiting (the three charging and discharging variables are all 0, )Any of the 4 states. When it is 0, it means that the electric vehicle is off-grid and cannot be charged or discharged.

[0194] Discharging is equivalent to the electric vehicle supplying power to the grid in reverse for the power grid; waiting means that the electric vehicle is parked in the parking lot without charging or discharging to the outside, but can be charged and discharged at any time; traveling means that the electric vehicle does not stop in the parking lot and drives normally outside.

[0195] The various states of electric vehicles are represented as follows:

[0196]

[0197] Among them, considering that electric vehicles have only one trip per day (one trip is counted between two charges of electric vehicles), the grid connection time and off-grid time are and And t off >t on , then the on-grid and off-grid status of electric vehicles The following constraints are met:

[0198]

[0199] in, When it is 1 or 0, it means that the electric vehicle is in grid-connected or off-grid state respectively.

[0200] When the time period t is greater than the grid-connected time and less than the off-grid time, 1 is the grid-connected state; when the period t is less than the grid-connected time and greater than the off-grid time, 0 means off-grid status.

[0201] In addition, the remaining power of electric vehicles must also meet battery capacity constraints, grid-connected remaining power constraints, and off-grid power demand constraints.

[0202] The battery capacity constraints are as follows:

[0203]

[0204] The constraints on off-grid surplus power demand are as follows:

[0205]

[0206] The grid-connected surplus power constraints are as follows:

[0207]

[0208] Among them, E EV,max and E EV,min The maximum and minimum storage capacity of electric vehicle batteries; The electricity demand for users’ travel; It refers to the remaining power after the user uses the electric vehicle. It is the remaining power of the electric vehicle when it is off the grid; It is the remaining power of electric vehicles when they are connected to the grid;

[0209] Formula (19) represents the remaining power of the electric vehicle, which needs to be greater than or equal to the minimum storage power of the electric vehicle battery and less than or equal to the maximum storage power of the electric vehicle battery. It is assumed that the power of the electric vehicle relative to the grid is 0 when it is off-grid.

[0210] Formula (20) indicates that the remaining power of the electric vehicle when it is off-grid needs to be greater than or equal to the user's travel power demand, which is set by the user based on their needs.

[0211] Formula (21) indicates that the remaining power of an electric vehicle when it is connected to the grid is the remaining power after the user uses it, which is obtained by analyzing the single charging data of the statistical charging pile as the historical basic data.

[0212] In actual operation, electric vehicle users adjust the time period t ad Benchmark operating (charging) power published by the dispatcher Assume that the benchmark operating (charging) power of electric vehicle i in time period t on a certain operating day released by the dispatching agency is Then in a certain adjustment period t ad The response power is A positive result indicates an upward response charge, and a negative result indicates a downward response charge. is the operating power of electric vehicle i during period t on a certain operating day.

[0213] (2) Air conditioning load response capability model.

[0214] Air conditioners have two modes: heating and cooling. The operating mechanism for the air conditioner load's regulation in cooling mode is theoretically the same as that in heating mode. The air conditioner load has two states: on and off. Adjustments to the operating state or power level can be made only in the on state, providing the grid with flexible output regulation. The on / off state of the air conditioner load is either "off" or "on" in the on state. "Off" means the air conditioner is in standby mode, not directly shutting down the power supply, while "on" indicates that the air conditioner is in cooling / heating operation.

[0215] In the startup state, the air conditioner is set according to the temperature T AC,setDelineation of temperature zones [T AC,min ,T AC,max ], under the premise that the indoor temperature is kept within the temperature range, the air conditioner switch is periodically paused. AC,min ,T AC,max are the minimum and maximum allowable values ​​of the indoor temperature, respectively. AC,set If set to 26℃, the temperature band [T AC,min ,T AC,max ]=[25℃,27℃], the specific value of the temperature band depends on the control strategy of the air-conditioning system and the user's comfort requirements.

[0216] The proportion of air conditioner operation time (air conditioner compressor working time) in a cycle is called "duty cycle". and Represent the time nodes of air conditioning startup and shutdown, t cut and t sta They represent the duration of the air conditioner being in standby state and running state during the period of periodic switching, P AC,R Indicates the rated cooling capacity of the air conditioner.

[0217] The duty cycle of an air conditioner refers to the ratio of the time the air conditioner compressor is on to the total usage cycle time during the air conditioner's usage cycle.

[0218] In this mode, air conditioning loads typically provide flexible power output to the grid through two methods: on / off control and temperature control. On / off control involves issuing an on / off command to the air conditioner, which then switches the air conditioner on or off. Temperature control, based on the human body's ability to sense changes in ambient temperature, has a delayed and tolerant nature. When the indoor temperature remains stable, the air conditioner's operating power is controlled by adjusting the temperature set point.

[0219] The equivalent thermal parameter (ETP) model is used to establish the thermodynamic model of air conditioning refrigeration. The relationship between its operating power and indoor and outdoor temperatures is as follows:

[0220]

[0221] Where C is the equivalent specific heat capacity; is the change in indoor temperature of air conditioner j in period t; is the cooling capacity of air conditioner j in period t; and represent the outdoor temperature and average indoor temperature of air conditioner j in period t respectively; R is the equivalent thermal resistance; is the energy efficiency ratio of air conditioner j; is the average operating power of air conditioner j in period t; is the “duty cycle” of air conditioner j in period t; is the rated power of air conditioner j. Compared with the power system dispatch cycle, the thermodynamic dynamic process of air conditioners is relatively short. After adjusting the set temperature, it will quickly run to a new steady state. Therefore, the time interval for air conditioner steady-state adjustment can be ignored.

[0222] According to formula (22), when the air conditioner is running in thermal steady state to keep the indoor temperature constant, the change of indoor temperature in period t is The value of is 0. At this time, the average operating power of the air conditioner is as follows:

[0223]

[0224] in, The linear coefficient representing the average operating power of the air conditioner and the indoor and outdoor temperature difference under thermal steady state; is a 0-1 variable that controls whether air conditioner j participates in the response in period t; The temperature set by the user in time period t; It is the adjustment amount for the user set temperature. Positive value indicates temperature increase and negative value indicates temperature decrease.

[0225] From formula (24), it can be seen that, assuming that the outdoor temperature remains unchanged for a certain period of time, the average operating power of the air conditioner can be directly controlled by adjusting the set temperature within the range allowed by the user's thermal comfort. Considering the user's tolerance range for indoor temperature changes is Then the operating power of air conditioner j after participating in the response satisfies the following constraints:

[0226]

[0227] in, These are the minimum and maximum tolerance ranges for indoor temperature changes by users; is the average operating power of the air conditioner.

[0228] After the user sets the tolerance range temperature, the air conditioner participates in the demand response, so the set tolerance temperature range will fluctuate up and down. For example, if the user sets 24℃, the user's tolerance temperature range is [23℃, 26℃]. are the lower and upper limits of the tolerance temperature range respectively.

[0229] In addition, prolonged indoor temperature changes can significantly affect user thermal comfort. Therefore, the response time of the air conditioning load is also subject to certain limitations, which are specifically achieved through the following constraints:

[0230]

[0231] Among them, t0 and t are the start and end time of the air conditioner temperature adjustment respectively; The number of time periods that the user allows air conditioner j to set the temperature adjustment; The number of time periods it takes for the air conditioner to adjust to the appropriate temperature. The physical meaning of formula (27) is the time it takes for the air conditioner to adjust to a comfortable temperature.

[0232] The operating power of the air-conditioning load is directly affected by user behavior data such as the start-up period, the set temperature adjustment range, the temperature adjustment time, and meteorological data such as the outdoor temperature. These factors also directly determine the responsiveness of the air-conditioning load. However, in the actual operation process, these user behavior data often have great uncertainty. In addition, the ambient temperature, humidity and other meteorological data will affect the responsiveness of the air-conditioning load by affecting the equivalent thermal resistance of the air, the temperature difference between indoors and outdoors, etc. For example, the benchmark operating power of air-conditioning j on a certain operating day released by the power grid dispatching agency is In the actual operation process, the actual operating power is reasonably arranged by comprehensively considering the electricity price and thermal comfort parameter conditions. The actual power consumption must meet the constraints of formulas (26)-(27). All power consumption curves within this range are the power consumption plans that can be actually executed by the user. Then, in a certain adjustment period t ad The response power is Positive is an upward response to the electric quantity, and negative is a downward response to the electric quantity. is the operating power of air conditioner j on a certain operating day.

[0233] A regulation period is the time interval during which demand-side resources are adjusted to achieve specific regulation objectives (such as peak regulation and frequency regulation) during power system operation. These periods are typically divided based on the operational requirements of the power system and the responsiveness of the resources. Regulation periods are divided according to different time scales. For example, hourly, minutely, or secondly regulation periods can be selected. A regulation cycle includes one or more regulation periods.

[0234] (3) Industrial load response capability model.

[0235] There are many types of industrial production, and there are many types of industrial loads with different properties. The response methods and response capabilities of different types of industrial loads are different to a certain extent.

[0236] Based on the characteristics of industrial production and different load types, industrial loads that can participate in demand response are divided into three types (continuous loads, transferable loads, and storage loads):

[0237] (a) Continuous loads. This type of load requires continuous operation and has a strong sequential nature in the production process. It requires interruption at the expense of higher response costs and often requires advance notification.

[0238] Assume that the load distribution of a continuous industrial load under normal production plan is The actual load distribution of the industrial load is It is expressed as follows (due to the high cost of interrupting production, only one interruption is considered within the scheduling cycle):

[0239]

[0240]

[0241] in, A 0-1 variable that controls whether load l participates in the response within the scheduling period t; is the load distribution of continuous industrial load l under normal production plan within the scheduling period t; is the intermediate variable involved in the response of load l; and They represent the start and end time periods of the load l interruption. Since only some processes can be interrupted and the interruption duration is limited, the following constraints are met:

[0242]

[0243] in, is the set of all interruptible periods of load l, and the interruption time corresponds to the production process that is interrupted at that time; and Respectively represent the minimum and maximum interruption duration.

[0244] (b) Transferable load. This type of load adopts an asynchronous parallel production mode and has flexible start-stop characteristics. By adjusting the production process, the power load can be transferred. The response capability model of transferable load is as follows:

[0245] Assume that the load distribution under the normal production plan of a certain transferable industrial load as follows:

[0246]

[0247] in, is the time series of the electricity load of the ith production process; is the starting period of the i-th (i=1, 2, 3, ..., n) production process when load l does not participate in the response; It is the end period of the last production process when load l does not participate in the response. If the order between the pth production process and the qth production process of the transferable industrial load is considered, it can be adjusted as follows:

[0248]

[0249]

[0250] in, A 0-1 variable that controls whether the i-th process between the original production processes p to q is in the k-th production process after adjustment; represents the time series of the electricity load of the kth production process after the original production process is adjusted; represents the time series of the adjusted electricity load of the original production process; Indicates that each load demand is connected. Among them, formula (22) means that only one production process can be executed at a time, and formula (36) means that each production process only needs to be executed once. On this basis, Modify as follows:

[0251]

[0252] in, The time series of the electricity load after the production process of load l is adjusted within the scheduling period. The transferable industrial load can also adjust the execution time of some production processes to transfer the load, as follows:

[0253]

[0254] in, is the actual load of load l in time period t; is the load of the original load sequence in period m, is a 0-1 variable indicating whether to transfer to time period t; n0 is the number of time periods in the scheduling cycle. t0 is the time to start participating in the response; Equations (25) to (28) aim to achieve load transfer by sorting the load values ​​of each time period in the scheduling cycle, while ensuring exist The sequence of the production process.

[0255] It means that the adjustable period should be between the start period of the i-th production process when load l does not participate in the response and the start period of the last production process when load l does not participate in the response.

[0256] (c) Storage loads. This type of load can shift electricity demand by completing production tasks ahead of or behind schedule. Storage-type industrial loads can provide regulated output to the grid by completing production tasks ahead of or behind schedule. Storage-type loads emphasize completing fixed production tasks within a specified timeframe. The response capability is modeled as follows:

[0257] Assume that the load distribution of a storage-type industrial load under normal production plan is The actual load distribution of the industrial load is as follows:

[0258]

[0259] in, and They represent the load l increasing or decreasing during the adjustment period t; and These are 0-1 variables that control whether the load l increases or decreases production tasks during the adjustment period t. Storage-type loads need to ensure that production tasks are completed within the scheduling cycle. Considering that the production tasks completed by their unit loads are consistent, the total load before and after the response must remain unchanged, as follows:

[0260]

[0261] In addition, storage loads also have maximum and minimum operating power restrictions as follows:

[0262]

[0263] in, and is the minimum and maximum operating power of load l.

[0264] From the above analysis, it can be concluded that the responsiveness of industrial load is determined by its own production nature, is directly related to the production plan, and is relatively fixed. Assume that the benchmark operating power of industrial load l released by the grid dispatching agency on a certain operating day is Then in a certain adjustment period t ad The response power is in, In a certain adjustment period t ad Actual operating power; positive indicates upward response power, negative indicates downward response power.

[0265] (4) User-side energy storage response capability model.

[0266] Behind-the-meter energy storage refers to energy storage equipment owned and used by end users (such as homes, commercial buildings, or industrial facilities). While meeting user needs, behind-the-meter energy storage provides grid-regulated output by adjusting the original charge and discharge schedule. For behind-the-meter distributed renewable energy + energy storage systems, energy storage is primarily used to store excess renewable energy generation during a specific period.

[0267]

[0268] in, are the power of user-side energy storage k in time periods t and t+1 respectively. Considering the remaining power of user-side energy storage k in the initial stage of the scheduling cycle is is the natural loss coefficient of energy storage k on the user side; is the operating power of the user-side energy storage k in time period t, where positive indicates charging of the energy storage device and negative indicates discharging of the energy storage device; and A 0-1 variable that controls the charging and discharging states of the user-side energy storage k in time period t; and is the rated charging power and rated discharging power of the user-side energy storage k; and The charging efficiency and discharging efficiency of the user-side energy storage k are also considered in the operation of the user-side energy storage:

[0269]

[0270] in, is the maximum storage capacity of user-side energy storage k; is the minimum remaining power of k.

[0271] The responsiveness of the user-side energy storage is determined by its own charging and discharging plan, maximum storage capacity, power usage demand, and remaining power at the initial stage of the cycle. Assume that the benchmark operating power of the user-side energy storage k released by the grid dispatching agency on a certain operating day is Then in a certain adjustment period t ad The response power is in, The user-side energy storage k is stored in a certain operation day during a certain adjustment period t ad The operating power of the device is positive; positive indicates upward response power, and negative indicates downward response power.

[0272] Step S2 builds a demand-side resource response model to quantify the responsiveness of various demand-side resources (such as electric vehicles, air conditioning loads, industrial loads, and user-side energy storage). This step provides key technical support for subsequent real-time rolling optimization, improving the regulation accuracy and efficiency of the power system, supporting the coordinated optimization of multiple resource types, and adapting to the real-time operational needs of the power system.

[0273] Step S3, specifically.

[0274] In actual operation, the response power of demand-side resources such as electric vehicles and air conditioning loads is greatly affected by meteorological data, which can lead to large deviations in the real-time response phase. This can cause regional demand-side resources to fail to respond to planned power, resulting in waste of power resources and uncertainty in power scheduling. This invention adopts MPC (Model Predictive Control) theory to perform real-time interactive prediction of demand-side resources.

[0275] Based on the concepts of MPC theory predictive control and rolling optimization, a demand-side resource rolling optimization model is constructed to continuously adjust the internal resource response plans to maximize the response power in the execution plan.

[0276] The real-time rolling response optimization solution of the demand-side resource response capability model includes:

[0277] With the goal of maximizing the power system regulation demand, the objective function and constraints of the real-time rolling response optimization solution are set; the planned load power of each demand-side resource in the regulation period t is used as the initial value of the demand-side response plan, as follows:

[0278]

[0279] Among them, P t r is the vector of each demand-side resource response plan in the regulation period t, n dr is the number of demand-side resources;

[0280] The demand-side resource response capability model is solved using the CPLEX solver. The adjustment period in the adjustment cycle of the operation day is t ad , from the adjustment period t ad -n, based on the latest operational day meteorological data, the response plan of each demand-side resource is dynamically adjusted every q hours; one adjustment cycle includes multiple adjustment periods;

[0281] During the rolling optimization process, the expected load power of each demand-side resource is within the elastic operating boundary, as follows:

[0282]

[0283] in, is the expected load power of demand-side resource i in regulation period t;

[0284] The model is solved every q hours to dynamically update the resource response plan for each demand side;

[0285] The response plan of each demand-side resource during the adjustment period is updated and sent synchronously to the demand-side user as follows:

[0286]

[0287] Until the adjustment period t ad The demand-side resource response plan has actually occurred, and the demand-side resources are adjusted during the adjustment period t ad The rolling optimization of the response plan has been completed;

[0288] In the first n hours of the next regulation period, the rolling optimization steps are repeated until the demand-side resource response plans for all regulation periods within the scheduling cycle have actually occurred, and the demand-side resource response behavior of the demand-side resources in the region in each regulation period within the regulation cycle is obtained.

[0289] During the real-time response phase, n hours before the regulation period, based on the latest meteorological data and the user behavior data of demand-side resource users during the regulation period predicted in real time by the correlation analysis model, a real-time response optimization model for regional demand-side resources is developed with the goal of maximizing the satisfaction of power system regulation needs. Based on the model's solution results, the real-time interaction strategy is dynamically adjusted, and the latest response plan is simultaneously released to each demand-side resource. Over the subsequent n hours, the system will repeat the above optimization solution steps every q hours, updating the response plan of each resource in real time to gradually narrow the actual response deviation of the aggregated resources as a whole. For example, q is 1 hour.

[0290] Assume that the first adjustment period of the operating day is t ad , then we need to start from time period t ad -n, the response plan of each demand-side resource is dynamically adjusted every hour based on the latest meteorological data.

[0291] The expected load power of each demand-side resource must be within the elastic operating boundaries of the demand-side resource on the operating day, as follows:

[0292]

[0293] in, is the expected load power of demand-side resource i in regulation period t; P i is the elastic operation boundary of demand-side resource i from the current moment to the end of the adjustment period;

[0294] The elastic operating boundary of the demand-side resources on the operating day is determined by the following steps:

[0295] Based on the latest operational day meteorological data at the current moment, the trained association analysis model is input to predict the demand-side user behavior data as follows;

[0296]

[0297] in, is the j-th user behavior data of demand-side resource i; Ψ(·) is the process of predicting user behavior data based on the latest operating day meteorological data by the association analysis model; MI k is the kth meteorological data corresponding to the user behavior data; is the meteorological data set that affects the behavior data of the jth user of demand-side resource i; Ω iis the user set of resource i on the demand side;

[0298] Based on the demand-side resource response capability quantification model and the predicted demand-side user behavior data, the upper and lower limits of the operating power of each demand-side resource within the adjustment period are calculated;

[0299]

[0300] where Φ(·) represents the constraint imposed by the quantitative model of the demand-side resource response capability on the load power of the demand-side resources in each period from the current moment to the end of the regulation period.

[0301] The calculated upper and lower limits of the operating power within the adjustment period are used as elastic operating boundaries and sent synchronously to the demand-side user to constrain the real-time response plan of the demand-side resources.

[0302] During the rolling optimization process, if the user behavior has already occurred due to other factors, According to the above solution results, the response plan of each demand-side resource in each adjustment period is updated and sent to the user synchronously, as follows:

[0303]

[0304] The objective function of this process is to maximize the regulation demand of the power system. Also includes changes in electricity costs for demand-side resource users In addition, since the demand side resources are in the regulation period t ad The response behavior will lead to changes in electricity consumption behavior in other time periods or even the entire scheduling cycle. The optimization range of the rolling optimization model is from the current time to the end of the scheduling cycle. The objective function solved by the real-time rolling response optimization is as follows:

[0305]

[0306] in, To meet the demand response capability of the power system in the region; R ad,t and F pun,t are the settlement income and deviation penalty fees in adjustment period t respectively; and C loss,i,t are the lth economic response cost and electricity satisfaction loss cost of demand-side resource i in regulation period t; I is the demand-side resource set.

[0307] The constraints of the real-time rolling response optimization solution include the elastic operating boundaries of demand-side resources, as well as electric vehicle, air-conditioning load, industrial load and user-side energy storage constraints.

[0308] The constraints of electric vehicles, air conditioning loads, industrial loads and user-side energy storage are the constraints in the electric vehicles, air conditioning loads, industrial loads and user-side energy storage response capability model in step S2.

[0309] Use the CPLEX solver to represent complex business problems as mathematical programming models to improve efficiency, rapidly implement strategies, and increase profitability.

[0310] The above optimization model is solved every 1 hour, and the response plan of each demand-side resource is dynamically updated until the adjustment period t ad The response behavior has actually occurred. So far, for each demand side resource readjustment period t ad The rolling optimization of the response plan has been completed. The rolling optimization steps are repeated n hours before the next regulation period until all regulation periods within the scheduling cycle have occurred. The final output is the demand-side resource response plan for the regulation period, which avoids waste of power resources and reduces transmission pressure.

[0311] For example, electric vehicle user A should respond with 50 kWh at 18:00 and 19:00, electric vehicle user B should respond with 30 kWh at 19:00, 20:00 and 21:00 respectively, air-conditioning user A should respond with 15 kWh at 18:00, and air-conditioning user B should respond with 10 kWh at 19:00 and 20:00 respectively.

[0312] The role of step S3 is to dynamically adjust the response plan of demand-side resources based on the latest meteorological data and user behavior forecasts through a real-time rolling optimization model to maximize the regulation needs of the power system and ensure the real-time and effectiveness of resource response.

[0313] In summary, the meteorological data-driven demand-side resource real-time rolling response optimization method according to the embodiment of the present invention has the following beneficial effects:

[0314] 1. By introducing a correlation analysis model based on meteorological data and user behavior data, this method can more accurately predict the responsiveness of demand-side resources. Compared with traditional methods, this method can adjust the response plan of demand-side resources in real time, reduce the deviation of demand-side resource response caused by changes in meteorological data, and improve the regulation accuracy of the power system based on accurate prediction driven by meteorological data.

[0315] 2. This invention dynamically adjusts the response plan of each demand-side resource within the regulation cycle by solving the demand-side resource response capability model in real time. This rolling optimization mechanism can timely update the demand-side resource response plan based on the latest meteorological data and user behavior data to ensure that the regulation needs of the power system are maximized.

[0316] 3. This invention introduces the concept of elastic operating boundaries for demand-side resources, ensuring that the expected load power of each demand-side resource fluctuates within a reasonable range. This constraint mechanism not only ensures the stability of the power system, but also improves the utilization efficiency of demand-side resources and avoids over-regulation or under-regulation.

[0317] 4. This invention comprehensively considers the responsiveness of various demand-side resources, including electric vehicles, air conditioning loads, industrial loads, and user-side energy storage, to construct a multi-type demand-side resource responsiveness model. By collaboratively optimizing these multiple demand-side resources, it maximizes flexibility and enhances the overall regulatory capacity of the power system.

[0318] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by any technician familiar with this technical field within the technical scope disclosed by the present invention should be covered by the scope of protection of the present invention.

Claims

1. A meteorological data-driven demand-side resource real-time rolling response optimization method, characterized in that: The steps include: Preprocessing the acquired meteorological data and corresponding user behavior data within a certain historical period of the area to be optimized to obtain a training sample set, and using the training sample set to train a trained association analysis model; inputting the acquired meteorological data during the adjustment period of the operating day into the trained association analysis model to obtain predicted demand-side user behavior data during the adjustment period; Obtaining corresponding demand-side user load based on the predicted demand-side user behavior data during the adjustment period, and building a demand-side resource responsiveness model based on the demand-side user load; Solving the demand-side resource response capability model in real time and in a rolling manner to obtain a demand-side resource response plan for the adjustment period; And perform real-time rolling solutions for each adjustment period in the adjustment cycle in turn to obtain the demand-side resource response plan of the demand-side resources in the region during the adjustment cycle.

2. The method according to claim 1, characterized in that The preprocessing of the acquired meteorological data within a certain historical period and the corresponding user behavior data in the area to be optimized to obtain a training sample set includes: Select user behavior data determined by meteorological data; Based on the selected user behavior data and corresponding meteorological data, data within an invalid period is eliminated to obtain meteorological data and corresponding user behavior data within a valid period as the first meteorological data and first user behavior data; wherein the valid period is a time period within the scheduling period during which user behavior data is sensitive to changes in meteorological data and has an actual response; Calculating the Pearson correlation coefficient between the first meteorological data and the first user behavior data, and selecting strongly correlated meteorological data and user behavior data with a correlation coefficient greater than or equal to a preset correlation threshold as the second meteorological data and the second user behavior data; Calculating the correlation between the operating day meteorological data and the second meteorological data using a grey correlation analysis method, and selecting meteorological data and corresponding user behavior data with a correlation greater than or equal to a preset correlation threshold as the third meteorological data and third user behavior data; Calculating a ratio of a distance from the third user behavior data to the sample center to an average distance, and removing third user behavior data having the ratio greater than a preset abnormality threshold as an outlier, thereby obtaining fourth meteorological data and fourth user behavior data from which the outliers have been removed, to form the training sample set; The fourth user behavior data includes continuity and time-based user behavior data.

3. The method according to claim 2, characterized in that Calculate the Pearson correlation coefficient between the first meteorological data and the first user behavior data Select The first meteorological data and the corresponding first user behavior data that are greater than or equal to the preset correlation threshold are used as the second meteorological data and the second user behavior data.

4. The method according to claim 3, characterized in that The association analysis model includes a multiple linear regression sub-model and a logistic regression sub-model; The fourth meteorological data is used as the independent variable and the fourth continuous user behavior data is used as the dependent variable to establish a multivariate linear regression sub-model; The fourth meteorological data is used as the independent variable and the corresponding fourth user behavior data of the time type is used as the dependent variable to establish a logistic regression sub-model.

5. The method according to claim 4, characterized in that: The association analysis model is trained using the training sample set to obtain a trained association analysis model, including: The least squares method was used to fit the coefficients in the multivariate linear regression sub-model, and the fitted coefficients were obtained as follows: in, is the coefficient of the fitted multiple linear regression sub-model; The cross entropy loss function is used as the cost function of the logistic regression sub-model to measure the data fitting degree of the logistic regression function. The loss function is as follows: Among them, J(τ) is the loss function; τ is the learning rate, Y f and Y r are the predicted value and actual value of user behavior data respectively; n H2 is the sample size of the time-based fourth user behavior data and the corresponding fourth meteorological data; The training sample set is used to iteratively optimize a and τ using the gradient descent method until the joint loss function of the multivariate linear regression sub-model and the logistic regression sub-model converges, thereby obtaining a trained association analysis model.

6. The method according to claim 1, characterized in that The expected load power of each demand-side resource must be within the elastic operating boundaries of the demand-side resource on the operating day, as follows: in, is the expected load power of demand-side resource i in regulation period t; P i is the elastic operation boundary of demand-side resource i from the current moment to the end of the adjustment period; The elastic operating boundary of the demand-side resources on the operating day is determined by the following steps: Based on the latest operational day meteorological data at the current moment, the trained association analysis model is input to predict the demand-side user behavior data as follows; in, is the j-th user behavior data of demand-side resource i; Ψ(·) is the process of predicting user behavior data based on the latest operating day meteorological data by the association analysis model; MI k is the kth meteorological data corresponding to the user behavior data; is the meteorological data set that affects the behavior data of the jth user of demand-side resource i; Ω i is the user set of resource i on the demand side; Based on the demand-side resource response capability quantification model and the predicted demand-side user behavior data, the upper and lower limits of the operating power of each demand-side resource within the adjustment period are calculated; where Φ(·) represents the constraint imposed by the quantitative model of the demand-side resource response capability on the load power of the demand-side resources in each period from the current moment to the end of the regulation period. The calculated upper and lower limits of the operating power within the adjustment period are used as elastic operating boundaries and synchronously sent to demand-side users to constrain the real-time response plan of demand-side resources.

7. The method according to claim 6, characterized in that The real-time rolling response optimization solution of the demand-side resource response capability model includes: With the goal of maximizing the power system regulation demand, the objective function and constraints of the real-time rolling response optimization solution are set; The planned load power of each demand-side resource in the regulation period t is used as the initial value of the demand-side response plan, as follows: Among them, P t r is the vector of each demand-side resource response plan in the regulation period t, n dr is the number of demand-side resources; The demand-side resource response capability model is solved using the CPLEX solver. The adjustment period in the adjustment cycle of the operation day is t ad , from the adjustment period t ad -n, based on the latest operational day meteorological data, the response plan of each demand-side resource is dynamically adjusted every q hours; one adjustment cycle includes multiple adjustment periods; During the rolling optimization process, the expected load power of each demand-side resource is within the elastic operating boundary, as follows: in, is the expected load power of demand-side resource i in regulation period t; The model is solved every q hours to dynamically update the resource response plan for each demand side; The response plan of each demand-side resource during the adjustment period is updated and sent synchronously to the demand-side user as follows: Until the adjustment period t ad The demand-side resource response plan has actually occurred, and the demand-side resources are adjusted during the adjustment period t ad The rolling optimization of the response plan has been completed; In the first n hours of the next regulation period, the rolling optimization steps are repeated until the demand-side resource response plans for all regulation periods within the scheduling cycle have actually occurred, and the demand-side resource response behavior of the demand-side resources in the region in each regulation period within the regulation cycle is obtained.

8. The method according to claim 7, characterized in that: The objective function solved by the real-time rolling response optimization is as follows: in, To meet the demand response capability of the power system in the region; R ad,t and F pun,t are the settlement income and deviation penalty fees in adjustment period t respectively; and C loss,i,t are the lth economic response cost and electricity satisfaction loss cost of demand-side resource i in regulation period t; I is the demand-side resource set.

9. The method according to claim 8, characterized in that The constraints of the real-time rolling response optimization solution include the elastic operating boundaries of demand-side resources, as well as electric vehicle, air-conditioning load, industrial load and user-side energy storage constraints.

10. The method according to any one of claims 1 to 9, characterized in that: The demand-side user loads include electric vehicles, air-conditioning loads, industrial loads and user-side energy storage loads; The demand-side resource response capability model includes electric vehicle, air conditioning load, industrial load and user-side energy storage response capability models.

Images