A temperature-sensitive load extraction method based on time-varying parameters of summer daytime temperature-sensitive load model

Abstract

The present application relates to a kind of summer day temperature-sensitive load model time-varying parameter-based temperature-sensitive load extraction method.The present application is based on summer medium-voltage load data and temperature data, establishes temperature-sensitive load model that can reflect the temperature change in day, introduces two assumptions to solve the problem of model identifiable, and proposes the matching data screening method and the two-stage identification method of temperature-sensitive load model time-varying parameter, obtains summer temperature-sensitive load model time-varying parameter by optimization solution, finally obtains the actual proportion of temperature-sensitive load and temperature-sensitive load curve.The method has good applicability, and better meets actual demand.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power systems, and particularly relates to a temperature-sensitive load extraction method based on time-varying parameters of a summer daytime temperature-sensitive load model. BACKGROUND

[0002] With the increasing proportion of new energy connected to the power grid, the fluctuation of the power supply side of the power system is increased, which further causes the difficulty of source-load balance of the power grid to rise. In order to maintain the balance between the source and the load, in recent years, the eyes of power system regulation have gradually turned to the load side, and more attention has been paid to flexible load which is easy to participate in dispatching. According to the latest statistics of some regional power systems, the proportion of temperature-sensitive load as a kind of flexible load in summer peak load can reach 40%. Therefore, it is of great significance to establish a temperature-sensitive load model, obtain the proportion of temperature-sensitive load, explore the regulation potential of temperature-sensitive load, realize the source-load balance of the power system, and ensure normal power supply at peak time. At present, there is no research on the time-varying temperature-sensitive load within a day, and the characteristics of temperature-sensitive load at different times in a day are not analyzed.

[0003] With the development of smart grid technology, many countries have deployed measurement devices in distribution networks, and available load measurement data has been expanded from high-voltage transmission stations to medium-voltage distribution stations, which greatly increases the amount of load data, especially the amount of load data similar to each other, providing a data basis for studying the influence of temperature change within a day on summer temperature-sensitive load. SUMMARY

[0004] In view of this, the application establishes a temperature-sensitive load model that can reflect the temperature change within a day based on medium-voltage load data and temperature data, introduces two assumptions to solve the problem of model identifiability, and proposes a matching data screening method and a two-stage identification method for time-varying parameters of the temperature-sensitive load model. Through optimization solution, the time-varying parameters of the summer temperature-sensitive load model, the actual proportion of the temperature-sensitive load, and the temperature-sensitive load curve are obtained. The method has good applicability and better meets the actual demand.

[0005] The application adopts the following scheme to realize it:

[0006] The application first provides a temperature-sensitive load extraction method based on time-varying parameters of a summer daytime temperature-sensitive load model, which includes the following steps:

[0007] Step (1): 96-point load data and temperature data of a user are collected, and data preprocessing is performed to weaken the influence of abnormal sampling points;

[0008] Step (2): Based on the load data obtained in step (1), the load is classified by K-means algorithm, and loads with similar shapes are classified into one class according to the size of Euclidean distance;

[0009] Step (3): establishing a temperature-sensitive load model, taking out a certain type of load from the load classification result of step (2), taking the sum of squared errors of the temperature-sensitive load model load calculation value and the load curve measured value as the optimization target, establishing an identification model of the temperature-sensitive load model full-time period parameter, and correcting the identification model according to the similarity of the load characteristics between the load data;

[0010] Step (4): based on the corrected identification model obtained in step (3), using a load screening method to screen the load data suitable for participating in identification as an identification cluster;

[0011] Step (5): based on the screened load data obtained in step (4), using a two-stage identification method to realize the solution of the model full-time period parameter and the discretization of the full-time period parameter respectively;

[0012] Step (6): taking the full-time period parameter and the discretized parameter obtained in step (5) as the basis to obtain the actual proportion of the temperature-sensitive load and the actual curve of the temperature-sensitive load.

[0013] According to the preferred scheme of the present application, step (3) is specifically: taking out the load curve of a type of user from the clustering result obtained in step (2), and taking N load curves therefrom, and establishing a corresponding temperature-sensitive load model for the N load curves;

[0014] wherein the original temperature-sensitive load model expression of the kth load curve at time t is:

[0015]

[0016] wherein, is the load value of the kth load curve at the sampling time t, the data sampling frequency is 15 minutes once, and there are 96 data points sampled in a day, so the range of t is 1-96, is the load value at the reference temperature T0, and respectively represent the proportion of temperature-sensitive load and non-temperature-sensitive load at the reference temperature T0, and have is the temperature of the kth load curve at time t, and T0 is the reference temperature;

[0017] The original temperature-sensitive load model shown in the above formula has three variables, namely and and has one equation constraint, that is, Therefore, the number of variables to be solved is 2, and the number of equations is 1; for N load curves, there are 96 sampling points in a day, the number of variables to be solved is 96*N*2, and the number of equations is 96*N; the number of variables is greater than the number of equations, therefore, there is a problem of model parameter identifiability for daily measurement data;

[0018] Given the large number of load curves, there must be K load curves with similar load characteristics. For such load curves, the following two assumptions are used to solve the problem of model parameter identifiability.

[0019] Assumption 1: For this type of load curve, their and The parameters are the same; at this time, and It can be uniformly represented as a t and b t ;

[0020] Assumption 2: For this type of load curve, their The parameters are proportional; at this point, Represented as load normalized value and proportionality coefficient R k The product of (k = 1, 2, ..., K), i.e.

[0021] After considering the two assumptions above, the relationship between load and temperature for the k-th load curve at time t is rewritten as follows:

[0022]

[0023] At this point, for K load curves, there are 96 sampling points per day, the number of variables to be solved is 96*2+K, and the number of equations is 96*K. As long as the number of equations involved in solving is greater than the number of variables to be solved, i.e. 96*K>96*2+K, i.e. K>3, and the sampling dates of the K loads are different, the problem of parameter identifiability can be solved.

[0024] With the optimization objective of minimizing the sum of squared errors between the calculated load values ​​and the measured load curve values ​​of the temperature-sensitive load model, the identification model for the full-time parameters of the modified temperature-sensitive load model is established as follows:

[0025]

[0026] ST

[0027]

[0028] in, Let be the load measurement value of the k-th load curve at time t. The load-temperature relationship at time t is obtained by the modified kth load curve The calculated load value.

[0029] According to a preferred scheme of the present application, the load screening method in step (4) is specifically divided into two steps: the first screening is based on the clustering result, and the load meeting hypothesis 2 is screened out; the second screening is further screening the load data meeting hypothesis 1 on the basis of the first screening, and the required load data set is formed.

[0030] Compared with the prior art, the present application has the beneficial effects including:

[0031] (1) The present application considers the intra-day temperature change, establishes a summer temperature-sensitive load model, and further extracts the summer temperature-sensitive load curve and the proportion through the time-varying parameters of the model;

[0032] (2) The present application introduces two hypotheses according to the similarity between the load data, overcomes the difficulty that the number of equations is less than the number of variables to be solved, and solves the problem of model identifiability;

[0033] (3) The present application constructs a data screening method based on the similarity between the load data and the hypothesis, and obtains a load data cluster suitable for model solution. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 The flow chart of the method of the present application;

[0035] Figure 2 The full-time period parameters and the discretized parameters obtained by solving under 0% disturbance;

[0036] Figure 3 The full-time period parameter solving results under 0%, 3%, 5% and 10% uniform disturbance applied to the constructed data;

[0037] Figure 4 The temperature-sensitive load proportion obtained by solving under 5% disturbance applied to the constructed data and the actual temperature-sensitive load proportion of the constructed data;

[0038] Figure 5 The temperature-sensitive load curve obtained according to the proportion obtained by solving in Figure 4 DETAILED DESCRIPTION

[0039] The present application will be further illustrated below in conjunction with the drawings and specific embodiments, and it should be understood that these embodiments are only used to illustrate the present application and not to limit the scope of the present application, and after reading the present application, those skilled in the art can make various equivalent modifications to the present application, which all fall within the scope defined by the appended claims of the present application.

[0040] ​The embodiment provides a temperature-sensitive load extraction method based on time-varying parameters of a summer daytime temperature-sensitive load model, as shown in the formula (1). Figure 1 The method comprises the following steps:

[0041] Step (1): 96-point load data and temperature data of a user are collected, and data preprocessing is performed to weaken the influence of abnormal sampling points;

[0042] Step (2): Based on the load data obtained in step (1), the K-means algorithm is used to classify the load, and loads with similar shapes are classified into one category according to the size of the Euclidean distance;

[0043] Step (3): A temperature-sensitive load model is established, a certain type of load is taken out from the load classification result in step (2), the sum of squared errors of the temperature-sensitive load model load calculation value and the load curve measured value is minimized as an optimization target, an identification model of the temperature-sensitive load model full-time period parameter is established, and the identification model is corrected according to the similarity of the load characteristics between the load data;

[0044] Step (4): Based on the corrected identification model obtained in step (3), a load screening method is used to screen the load data suitable for participating in identification as an identification cluster;

[0045] Step (5): Based on the screened load data obtained in step (4), a two-stage identification method is used to realize the solution of the model full-time period parameter and the discretization of the full-time period parameter;

[0046] Step (6): The full-time period parameter and the discretization parameter obtained in step (5) are used as the basis to obtain the actual proportion of the temperature-sensitive load and the actual curve of the temperature-sensitive load.

[0047] The step (1) is specifically that, for the load curve P of each 10kV user, the following smoothing processing is performed:

[0048] For the first and 96th time, no processing is performed;

[0049] For the second and 95th time, the processing is as follows:

[0050]

[0051]

[0052] For the third to 94th time, the processing is as follows:

[0053]

[0054] In the above formula, L t is the original load value at the t time, and L' tThe load value at time t after smoothing processing;

[0055] The load value is normalized to 0 to 1, and the specific formula is:

[0056]

[0057] Wherein, L" is the normalized load value, L' min is the minimum load value of the load curve P after smoothing processing, L' max is the maximum load value of the load curve P after smoothing processing.

[0058] The step (2) is specifically classified by K-means algorithm. As a classic algorithm in clustering analysis field, the basic idea of K-means algorithm is to divide N data points into K categories, so that the distance from the cluster center of each category to all data points in the category is minimized.

[0059] The implementation steps of K-means algorithm clustering once are as follows

[0060] 1) Randomly select K load curves as initial cluster centers;

[0061] 2) Calculate the Euclidean distance from N load curves to K cluster centers one by one, and divide the data point into the category where the cluster center is the smallest;

[0062] 3) After dividing N load curves, calculate the average value of data points in K categories as the new cluster center of K categories;

[0063] 4) Repeat steps 2) and 3) until the cluster centers of K categories do not change any more.

[0064] Two evaluation indexes are used to determine the best cluster number and the best cluster result considering the randomness of initial cluster center. First, the classification accuracy index (I DB ) is used to determine the best cluster number. I DB The smaller the value is, the smaller the intra-class distance is, the larger the inter-class distance is, and the more significant the classification effect is. The calculation method of I DB is

[0065]

[0066] In the above formula: k is the cluster number; c i , c j are the cluster centers of the i-th and j-th categories respectively; are the average distances from the load curves in the i-th and j-th categories to the cluster centers c i , c j of the corresponding categories respectively.

[0067] The error sum of squares index (I SSE ) is used to evaluate the clustering results corresponding to different initial clustering centers. First, the number of clustering is set, and then the I SSE value of each clustering result is calculated, and finally the clustering result with the minimum I SSE is selected. The I SSE calculation method is as follows

[0068]

[0069] In the above formula, n i is the number of load curves in the i-th category; c ij is the j-th load curve in the i-th category.

[0070] According to the above two indexes, the best clustering result of the load is obtained through multiple clustering.

[0071] The step (3) is specifically: taking out the load curve of a user from the clustering result obtained in step (2), taking N load curves therefrom, and establishing a corresponding temperature-sensitive load model for the N load curves.

[0072] In the above formula, t is the time of the k-th load curve, and the original temperature-sensitive load model expression of the k-th load curve at time t is as follows

[0073]

[0074] In the above formula, L is the load value of the k-th load curve at the sampling time t, the data sampling frequency used in the method is 15 minutes, and there are 96 data points in a day, so the range of t is 1-96, is the load value at the reference temperature T0, and represent the proportion of temperature-sensitive load and non-temperature-sensitive load at the reference temperature T0, and have is the temperature of the k-th load curve at time t, and T0 is the reference temperature, which is taken as 30℃.

[0075] The original temperature-sensitive load model shown in the above formula has three variables, i.e. and and one equation constraint, i.e. Therefore, the number of variables to be solved is 2, and the number of equations is 1; for N load curves, there are 96 sampling points in a day, the number of variables to be solved is 96*N*2, and the number of equations is 96*N; the number of variables is greater than the number of equations, so there is a problem of model parameter identifiability for single-day measurement data.

[0076] Given the large number of load curves, there must be K load curves with similar load characteristics. For such load curves, the following two assumptions are used to solve the problem of model parameter identifiability.

[0077] Assumption 1: For this type of load curve, their and The parameters are the same; at this time, and It can be uniformly represented as a t and b t .

[0078] Assumption 2: For this type of load curve, their The parameters are proportional; at this point, This can be expressed as a load normalization value. and proportionality coefficient R k The product of (k = 1, 2, ..., K), i.e.

[0079] The rationale for Assumption 1 is that loads with similar load types and load curve shapes tend to exhibit similar electricity consumption behaviors; that is, during temperature changes, temperature-sensitive loads will also exhibit similar changes. Therefore, and They may be similar or even identical.

[0080] Assumption 2 is reasonable because loads with similar load types and load curve shapes have highly similar normalized load curves, which makes it possible for these load values ​​to have an approximate proportional relationship. Therefore, the loads in the temperature-sensitive load model are also very likely to exhibit proportional properties.

[0081] After considering the two assumptions above, the relationship between load and temperature at time t for the k-th load curve can be rewritten as follows:

[0082]

[0083] At this point, for K load curves, there are 96 sampling points per day, the number of variables to be solved is 96*2+K, and the number of equations is 96*K. As long as the number of equations involved in the solution is greater than the number of variables to be solved, i.e., 96*K>96*2+K, or K>3, and the sampling dates of the K loads are different (to ensure that the temperature data corresponding to these loads are different), the problem of parameter identifiability can be solved. In addition, including other similar loads in the solution object improves the practicality and universality of the parameters.

[0084] With the optimization objective of minimizing the sum of squared errors between the calculated load values ​​and the measured load curve values ​​of the temperature-sensitive load model, the identification model for the full-time parameters of the modified temperature-sensitive load model is established as follows:

[0085]

[0086] S.T.

[0087]

[0088] The step (4) is specifically that based on the two assumptions in the step (3) and the obtained corrected identification model, a supporting load screening method is proposed to screen the load data meeting the two assumptions and suitable for participating in identification as an identification cluster;

[0089] The screening method is specifically divided into two steps: the first screening is based on the clustering result to screen out the load meeting the assumption 2; the second screening is further screening the load data meeting the assumption 1 on the basis of the first screening to form the load data set required for identification. With the load k as the main load for identification, the remaining load to be screened needs to be compared with the load k, and the load screening method specifically includes the following steps:

[0090] 1) The first screening

[0091] If the load i and the load k meet the assumption 2, it can be known that the and should show proportional properties, for the load k, from and it can be known that can be expressed as:

[0092]

[0093] At the same time, based on the clustering result and the theory of the assumption 1, it is considered that the loads of the same class approximately have the same and Therefore, the ratio between the of the load i and the load k at the time t can be expressed as:

[0094]

[0095] Among them,

[0096] For the load meeting the assumption 2, the at the 96 sampling points should be the same, that is, the variance is 0; since the specific values of a t and b t are unknown, the accurate value of cannot be determined, but its interval can be estimated;

[0097] The estimation of the interval is considered as a interval estimation of a​t and b t For a variable optimization problem, the goal is... Minimize and maximize the value, with constraint a t +b t =1, 0≤a t ≤1 and 0≤b t ≤1; for time t, respectively at as well as To obtain the maximum value, we can get... The interval is:

[0098]

[0099] Then combine It can be estimated The interval:

[0100]

[0101] Based on the above results, The value of the midpoint of the interval is used as The estimated value:

[0102]

[0103] For each of the 96 sampling points, the estimated value was obtained using the same screening method described in the first step. Then 96 sampling points were obtained. Composed of sequences Calculate the variance of the sequence, denoted as δ. i,k ; δ for all loads in the load cluster where load k is located i,k Sort, δ i,k The smaller the value, the more it conforms to Assumption 2, meaning it is more suitable as an identification cluster for load k; δ is then selected. i,k The minimum n1 loads yield the dataset S. refine1 For further screening;

[0104] 2) Second screening

[0105] Dataset S obtained through coarse screening refine1 It may not fully satisfy Assumption 1, so further analysis of the dataset S that satisfies Assumption 2 is needed. refine1 The dataset was filtered to obtain a load dataset that better fits Hypothesis 1;

[0106] If load i and load k satisfy Assumption 1, then their and They should be the same, but because and The specific numerical values ​​are unknown and cannot be directly compared; therefore, an indicator is defined. Indirect comparison and

[0107]

[0108] in, Source This expression is essentially a linear function. And there are

[0109] Due to dataset S refine1 The load in the middle satisfies assumption 2, so it can be known that and It exhibits a proportional property, therefore if the load i satisfies assumption 1, then and Similarly, it should exhibit proportional properties, so 96 sampling points The variance should be the same, i.e., the variance is 0, so we use the same method. The variance was used as the basis for data selection.

[0110] Obtained by estimation approximation

[0111]

[0112] in, and These represent the load change and temperature change of load i at sampling point t on day d, respectively, obtained from the load and temperature data of load i on day d and day d-1; This represents the load value of load i at sampling point t on day d. This represents the temperature value corresponding to sampling point t of load i on day d; d and D in the formula both represent dates, and the use of uppercase and lowercase indicates that the dates of load i and load k are not the same.

[0113] For all 96 sampling points, the estimated values ​​were obtained using the second screening method described above. Then 96 sampling points were obtained. Composed of sequences Calculate the variance of the sequence, denoted as δ. Aik For dataset S refine1 δ of all loads Aik Sort, δ Aik The smaller the value, the more it conforms to Assumption 1, meaning it is more suitable as an identification cluster for load k; δ is then selected. Aik The minimum n² loads yield the dataset S. refine2 It is used for parameter identification of temperature-sensitive load models.

[0114] The step (5) is specifically, from the screened load data obtained in step (4), a two-stage identification method is proposed to solve the model full-period parameters and the discretization of the full-period parameters respectively:

[0115] 1) The first stage full-period parameter identification

[0116] In the first stage of the two-stage identification method, the parameters a t , b t , and R k are identified.

[0117] The entire identification process can be regarded as an optimization problem, which aims to find suitable parameters a t , b t , and R k , so that the deviation between the measured power and the calculated power based on is minimized, and the optimization problem can be expressed as:

[0118]

[0119] S.T.

[0120]

[0121] The fmincon function of the matlab platform is used to solve the above optimization problem to obtain the full-period parameters of the summer temperature-sensitive load model.

[0122] 2) The second stage of the discrete parameter identification

[0123] The second stage of the two-stage identification method is the discretization of the full-period parameters of the summer temperature-sensitive load model. The values of the 96 points of the full-period parameters are not the same, and if the 96 points of the model parameter data are saved for each load or each load group, there will be inconvenience in data storage or analysis and calculation. The discretized parameters better retain the characteristics of the full-period parameters in each section, and also solve the practicality problem of the full-period parameters, and can suppress the randomness of the full-period parameters caused by load random fluctuations and measurement noise;

[0124] Assume that there are N sections, where N is usually less than 5, because the daily load curve generally contains one to two peak periods, such as the noon peak and the evening peak, and several rapid transition periods, such as the morning load rising period and the evening load falling period, and the sum of the number of these periods is usually less than 5. At the same time, there will be N-1 section points, denoted as n=[n1,…n N-1 ] and N groups of discretized parameters and (i = 1, 2, …, N);

[0125] The all-time period parameter is denoted as wherein and are the all-time period parameters of the summer temperature-sensitive load model, which are composed of and ; the discretization parameter is denoted as wherein and are the discretization parameters of the summer temperature-sensitive load model; for the form of the all-time period parameter is the same as that of the discretization parameter;

[0126] The discretization process can also be regarded as an optimization problem, the objective function of which is to minimize the deviation between the all-time period parameter and the discretization parameter, and the optimization problem can be expressed as:

[0127]

[0128] S.T.

[0129]

[0130] wherein represents the Euclidean distance between the all-time period parameter and the discretization parameter, and the optimization problem is solved by using the OPTI toolbox of the matlab platform to obtain the discretization parameter of the summer temperature-sensitive load model.

[0131] The all-time period parameter identification and the discretization parameter identification are combined to form a two-stage parameter identification method. A civil load cluster is constructed for solving, which contains 6 load curves and 6 days of temperature data, Figure 2 the all-time period parameter solved under 0% perturbation is shown and the discretization parameter is Figure 3 The all-time period parameter solving results under 0%, 3%, 5%, and 10% uniform perturbation applied to the constructed data are shown, and it can be seen from the figure that in the case of no perturbation, the parameter solving result is exactly the same as the actual value, and after the perturbation, the solving result fluctuates but is still relatively close to the actual result in general, which shows that the identification method not only realizes the solution of the model parameter, but also has good robustness.

[0132] The step (6) is specifically that, by taking the all-time period parameter and the discretization parameter obtained in the step (5) as the basis, the actual proportion of the temperature-sensitive load and the actual curve of the temperature-sensitive load can be obtained.

[0133] In the temperature-sensitive load model The physical meaning of the ratio of the temperature-sensitive load and the non-temperature-sensitive load at the reference temperature T0 is the temperature-sensitive load and the non-temperature-sensitive load, respectively, and And The physical meaning of the ratio of the temperature-sensitive load and the non-temperature-sensitive load at the reference temperature T0 is the temperature-sensitive load and the non-temperature-sensitive load, respectively, and However, it is actually impossible for the temperature to be T0 at 96 points in a day. The temperature The difference from the reference temperature T0 will change the ratio of the temperature-sensitive load and the total load, and the ratio of the temperature-sensitive load becomes And the sum of the ratio of the temperature-sensitive load and the non-temperature-sensitive load Is no longer equal to 1, thereby causing a change in the total load. Therefore, the ratio of the temperature-sensitive load in the total load at time t of the kth load curve needs to be recalculated in combination with the actual temperature data:

[0134]

[0135] Wherein, The actual temperature-sensitive load ratio of the load k at time t, Figure 4 Showed the temperature-sensitive load ratio obtained by solving under the application of 5% disturbance to the constructed data and the actual temperature-sensitive load ratio of the constructed data;

[0136] After obtaining the actual ratio , multiplying it by the actual load curve obtains the actual curve of the temperature-sensitive load, Figure 5 Showed the temperature-sensitive load curve obtained from the ratio obtained by solving and the actual ratio. From Figure 4 The comparison of the solving results of Figure 4 And Figure 5 With the actual results can be seen that even in the case of disturbance, the temperature-sensitive load ratio and the temperature-sensitive load curve solved have a small difference from the actual values, and it can be seen that the method not only can realize the extraction of the temperature-sensitive load component in the load curve, but also has good accuracy and robustness.

[0137] The above only describes the preferred embodiments of the present application, and any changes and modifications made within the scope of the patent application of the present application shall be covered by the scope of the present application.

Claims

1. A method for extracting temperature-sensitive loads based on time-varying parameters of a summer intraday temperature-sensitive load model, characterized in that... Includes the following steps: Step (1): Collect 96 load and temperature data points from users and perform data preprocessing to reduce the impact of abnormal sampling points; Step (2): Based on the load data obtained in step (1), the load is classified by the K-means algorithm, and loads with similar shapes are grouped into one category according to the Euclidean distance. Step (3): Establish a temperature-sensitive load model. Take out a certain type of load from the load classification results in step (2). With the goal of minimizing the sum of square errors between the calculated load value and the measured load curve value of the temperature-sensitive load model, establish an identification model for the parameters of the temperature-sensitive load model for all time periods. And correct the identification model based on the similarity of load characteristics between load data. Step (4): Based on the corrected identification model obtained in step (3), a load screening method is used to screen load data suitable for identification clusters to participate in identification. Step (5): Based on the filtered load data obtained in step (4), a two-stage identification method is adopted to solve the model parameters for the entire time period and to discretize the parameters for the entire time period. Step (6): Based on the full-time parameters and discretized parameters obtained in step (5), the actual proportion of temperature-sensitive load and the actual curve of temperature-sensitive load are obtained.

2. The method for extracting temperature-sensitive loads based on time-varying parameters of a summer intraday temperature-sensitive load model according to claim 1, characterized in that: Specifically, step (1) involves smoothing the load curve P of each 10kV user as follows: No processing is performed for time points 1 and 96; For time points 2 and 95, the processing is as follows: ； ； For time points 3 through 94, the processing is as follows: ； In the above formula, The original load value at time t. The load value at time t is the smoothed value; The load is then normalized to a value between 0 and 1. The specific formula is as follows: in, This is the normalized load value. The minimum load value of the load curve P after smoothing. This represents the maximum load value of the load curve P after smoothing.

3. The method for extracting temperature-sensitive loads based on time-varying parameters of a summer intraday temperature-sensitive load model according to claim 1, characterized in that: The specific steps (3) are as follows: Take the load curve of a type of user from the clustering results obtained in step (2), take N load curves from them, and establish a corresponding temperature-sensitive load model for these N load curves. The original temperature-sensitive load model expression for the k-th load curve at time t is: in, Let be the load value of the k-th load curve at sampling time t. The data sampling frequency is once every 15 minutes, and 96 data points are sampled per day. Therefore, the range of t is 1-96. At the reference temperature The load value below, and These represent the reference temperature. The ratio of temperature-sensitive loads to non-temperature-sensitive loads, and has , Let be the temperature of the k-th load curve at time t. The reference temperature is ; The original temperature-sensitive load model shown in the above equation has three variables, namely... , and And there exists one equality constraint, namely Therefore, the number of variables to be solved is 2, and the number of equations is 1; for N load curves, there are 96 sampling points in a day, the number of variables to be solved is 96*N*2, and the number of equations is 96*N; the number of variables is greater than the number of equations, therefore, there is a problem of model parameter identifiability for daily measurement data; Given the large number of load curves, there must be K load curves with similar load characteristics. For such load curves, the following two assumptions are used to solve the problem of model parameter identifiability. Assumption 1: For this type of load curve, their and The parameters are the same; at this time, and Unified representation as and ; Assumption 2: For this type of load curve, their The parameters are proportional; at this point, Represented as load normalized value and proportionality coefficient The product of, i.e. ; After considering the two assumptions above, the relationship between load and temperature for the k-th load curve at time t is rewritten as follows: At this point, for K load curves, there are 96 sampling points per day, the number of variables to be solved is 96*2+K, and the number of equations is 96*K. As long as the number of equations involved in solving is greater than the number of variables to be solved, i.e. 96*K>96*2+K, i.e. K>3, and the sampling dates of the K loads are different, the problem of parameter identifiability can be solved. With the optimization objective of minimizing the sum of squared errors between the calculated load values ​​and the measured load curve values ​​of the temperature-sensitive load model, the identification model for the full-time parameters of the modified temperature-sensitive load model is established as follows: in, Let be the load measurement value of the k-th load curve at time t. To obtain the relationship between load and temperature at time t using the corrected k-th load curve. The calculated load value.

4. The method for extracting temperature-sensitive loads based on time-varying parameters of a summer intraday temperature-sensitive load model according to claim 3, characterized in that: The load screening method in step (4) is specifically divided into two steps: the first screening is based on the clustering results to screen out the loads that meet hypothesis 2; the second screening is based on the first screening to further screen the load data that meet hypothesis 1 to form the load dataset required for identification.

5. The method for extracting temperature-sensitive loads based on time-varying parameters of a summer intraday temperature-sensitive load model according to claim 4, characterized in that: Using load k as the primary load for identification, all other loads to be screened must be compared with load k. The load screening method specifically includes the following steps: 4.1) First screening If load i and load k satisfy assumption 2, then it can be known that... and It should exhibit proportional properties, for load k, by ,and It can be known that It can be represented as: Meanwhile, based on the clustering results and the theory of Hypothesis 1, it is believed that loads of the same type have approximately the same... and Then the load i and the load k at time t and The ratio between It can be represented as: in , ; For a load that satisfies Assumption 2, its sampling points are... They should be the same, i.e., the variance is 0; since we don't know... and The exact value cannot be determined. It has an exact value, but its range can be estimated; The estimation of the interval is regarded as a... and For a variable optimization problem, the goal is... Minimize and maximize the value, with constraints as follows , as well as For time t, respectively at as well as To obtain the maximum value, we can get... The interval is: Then combine , estimate The interval: Based on the above results, The value of the midpoint of the interval is used as The estimated value: For all 96 sampling points, the estimated values ​​were obtained using the method described in step 4.1). Then 96 sampling points were obtained. Composed of sequences Calculate the variance of the sequence, denoted as . For all loads in the load cluster where load k is located... Sort them. The smaller the value, the more it conforms to Assumption 2, meaning it is more suitable as an identification cluster for load k; screening... smallest A load is obtained to get the dataset. For further screening; 4.2) Second screening Dataset obtained through coarse screening It may not fully satisfy Assumption 1, so further analysis of the dataset that satisfies Assumption 2 is needed. The dataset was filtered to obtain a load dataset that better fits Hypothesis 1; If load i and load k satisfy Assumption 1, then their and They should be the same, but because and The specific numerical values ​​are unknown and cannot be directly compared; therefore, an indicator is defined. Indirect comparison and : in, Source This formula is essentially a linear function. And there are ; Due to the dataset The load in the middle satisfies assumption 2, so it can be known that and It exhibits a proportional property, therefore if the load i satisfies assumption 1, then and Similarly, it should exhibit proportional properties, so 96 sampling points The variance should be the same, i.e., the variance is 0, so we use the same method. The variance was used as the basis for data selection; Obtained by estimation approximation : in, and These represent the load change and temperature change of load i at sampling point t on day d, respectively, obtained from the load and temperature data of load i on day d and day d-1; This represents the load value of load i at sampling point t on day d. This represents the temperature value corresponding to sampling point t of load i on day d; d and D in the formula both represent dates, and the use of uppercase and lowercase indicates that the dates of load i and load k are not the same. For all 96 sampling points, the estimated values ​​were obtained using the method described in step 4.2). Then 96 sampling points were obtained. Composed of sequences Calculate the variance of the sequence, denoted as . ; For dataset All loads Sort them. The smaller the value, the more it conforms to Assumption 1, meaning it is more suitable as an identification cluster for load k; screening... smallest A load is obtained to get the dataset. It is used for parameter identification of temperature-sensitive load models.

6. The method for extracting temperature-sensitive loads based on time-varying parameters of a summer intraday temperature-sensitive load model according to claim 5, characterized in that: Step (5) specifically involves: 5.1) First stage: Full-time parameter identification In the first stage of the two-stage identification method, the parameters are identified. , , and ; The entire identification process can be viewed as an optimization problem, the goal of which is to find suitable parameters. , , and This enables the measurement of power. and based on Calculated power The optimization problem, which aims to minimize the deviation between the two, is expressed as: Solve the above optimization problem to obtain the parameters of the summer temperature-sensitive load model for all time periods; 2) Second stage discrete parameter identification The second stage of the two-stage identification method is the discretization of the parameters of the summer temperature-sensitive load model throughout the entire period; Suppose there are N segments, and N-1 segmentation points, denoted as... and N sets of discretization parameters and ; The full-time parameter is denoted as ,in and The parameters for the summer temperature-sensitive load model throughout the entire period are derived from... and Composition; the discretization parameter is denoted as ,in and These are the parameters of the discretized summer temperature-sensitive load model; for , , The form is the same as it; The discretization process can also be viewed as an optimization problem, the objective function of which is to minimize the deviation between the full-time parameters and the discretized parameters. This optimization problem can be expressed as: in The Euclidean distance between the full-time parameters and the discretized parameters is used to solve this optimization problem to obtain the discretized parameters of the summer temperature-sensitive load model.

7. The method for extracting temperature-sensitive loads based on time-varying parameters of a summer intraday temperature-sensitive load model according to claim 1, characterized in that: Step (6) specifically involves: Based on actual temperature data, recalculate the proportion of temperature-sensitive load in the total load for the k-th load curve at time t: in, This represents the actual proportion of temperature-sensitive load k at time t. After obtaining the actual percentage Then, multiply it by the actual load curve to obtain the actual curve of the temperature-sensitive load.

Images