A method for evaluating the regulation capability of adjustable loads based on power grid regulation needs

Abstract

The present invention discloses a method for evaluating the regulation capability of an adjustable load oriented to the demand for power grid regulation, and belongs to the technical field of power grid demand side management. A generalized model of an adjustable load is constructed based on the resource characteristics of the demand side, and the quadratic regression dynamic relationship between energy consumption behavior and a variety of uncertain factors is fitted. The clustering method is used to extract similar energy consumption behaviors between different users. The quadratic regression parameters of the energy consumption behavior and uncertain factors of each user in the source domain and the target domain are obtained based on the clustering results. The cosine similarity of parameter features between users in the same class is calculated. The probability distribution of the adjustable load regulation capability of the target domain is obtained based on the cosine similarity. The energy consumption behavior characteristics of users in the source domain and the target domain are aligned based on the local maximum mean difference. The online transfer learning network model based on feature alignment is used to evaluate the adjustable load regulation capability of the target domain. The present invention can realize the evaluation of the adjustable load regulation capability of the demand side, and provide theoretical reference and technical support for demand side management.

Description

Technical Field

[0001] The present invention belongs to the technical field of power grid demand side management, and in particular relates to a method for evaluating adjustable load regulation capability oriented to power grid regulation demand. Background Art

[0002] As the proportion of distributed resources connected to the grid increases, a large number of traditional units are retired, and a large number of flexible loads are connected in an unordered manner on the demand side. Peak loads are becoming increasingly prominent, and distributed power output cannot track the load curve. Wind and solar power curtailment are serious, and the grid's regulation capacity is seriously insufficient. Safe and stable operation faces challenges, which can easily lead to grid overload and even collapse, resulting in frequent major blackouts. Implementing demand response on the demand side is a key direction for improving system regulation capabilities, promoting the absorption of new energy, and ensuring system safety and stability.

[0003] In light of the new power system's operational characteristics, it is urgent and necessary to incorporate demand-side adjustable load resources into normalized grid operation and regulation, promote the grid's transformation from "source follows load" to "source-load interaction," and implement an assessment of the adjustable load's regulation capacity. To address the source-load imbalance problem caused by the grid's "double high" characteristics, demand-side adjustable load resources are urgently needed to be explored. Therefore, an assessment method for the adjustable load regulation capacity tailored to grid regulation needs is of great research significance. Summary of the Invention

[0004] The purpose of the present invention is to provide an adjustable load regulation capability evaluation method for grid regulation needs, so as to incorporate demand-side adjustable load resources into normal grid operation regulation and control, and promote the transformation of the grid from "source follows load" to "source-load interaction".

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for evaluating the regulation capability of adjustable loads oriented to power grid regulation needs includes:

[0007] Obtaining energy consumption data of users to be evaluated on the demand side, wherein the user energy consumption data includes user energy consumption behavior, uncertainty factors, and electricity prices;

[0008] The energy consumption data of the user to be evaluated is input into a trained online transfer learning network model, and the model outputs a user-adjustable load adjustment ratio.

[0009] Furthermore, the online transfer learning network model is trained by the following method:

[0010] Obtain historical energy consumption data of the source domain and the target domain;

[0011] A source domain adjustable load generalized model and a target domain adjustable load generalized model are constructed based on user energy consumption behaviors and uncertain factors affecting user energy consumption behaviors in source domain and target domain data, respectively. Based on the source domain adjustable load generalized model and the target domain adjustable load generalized model, quadratic regression dynamic relationships between source domain and target domain user energy consumption behaviors and multiple uncertain factors are obtained by fitting.

[0012] Clustering the historical energy consumption data of the source and target domains is performed to extract similar energy consumption behaviors among different users. Based on the clustering results of the source and target domains and the quadratic regression dynamic relationship between the energy consumption behaviors of the source and target domain users and various uncertain factors, the quadratic regression parameters of the energy consumption behaviors of the source and target domain users within the class and the uncertain factors are obtained.

[0013] Calculate the cosine similarity of the quadratic regression parameter characteristics of the target domain and source domain users' energy consumption behavior and uncertain factors within the class, and obtain the probability distribution of the adjustable load regulation capacity of the target domain based on the cosine similarity;

[0014] Based on the probability distribution of the adjustable load regulation capability of the target domain, the energy consumption behavior characteristics of users in the source domain and the target domain are aligned according to the local maximum mean difference, and the loss function is minimized by continuously training the model until the model converges.

[0015] Furthermore, the generalized model of the adjustable load is:

[0016] P t =P b,t (R m )+ΔP t (ΔR)=P b,t (R m )(1+f t (ΔR)

[0017] Where, P t is the adjustable load at time t; P b,t (R m ) is the adjustable load baseline load at time t, which is related to the uncertainty factor R that affects the user's energy consumption behavior. m Related; ΔP t (ΔR) is the adjustable load change value at time t, which is related to the uncertain factor ΔR that affects the user's energy consumption behavior; R m and ΔR are the average value and change value of the uncertainty factor in a certain period of time respectively; f t (ΔR) is P b,t (R m ) and ΔP t (ΔR) function.

[0018] Furthermore, the quadratic regression dynamic relationship between the user's energy consumption behavior and multiple uncertain factors is:

[0019] f t (ΔR)=f 1,t +f 2,t ΔR+f 3,t ΔR 2

[0020] Where, f 1,t 、f 2,t 、f 3,t are the parameters of the quadratic regression.

[0021] Furthermore, clustering the historical energy usage data of the source domain and the target domain to extract similar energy usage behaviors among different users includes:

[0022] Randomly select K users’ energy consumption history data as the initial cluster centers;

[0023] Calculate the distance between the remaining samples and the cluster center and assign the samples to the nearest cluster:

[0024]

[0025] Where K is the number of clusters, v ik Indicates whether the i-th user sample belongs to the k-th category, v ik 1 means it belongs to class k, v ik 0 means it does not belong to class k, d(c m ,x i ) is the sample x i To cluster center c m distance;

[0026] Update cluster centers:

[0027]

[0028] Where N is the number of users;

[0029] Determine the convergence condition to minimize the following formula:

[0030]

[0031] Furthermore, the cosine similarity of the quadratic regression parameter features of the energy consumption behaviors and uncertain factors of the target domain and source domain users within the class is calculated according to the following steps:

[0032] Assume that the quadratic regression parameter set of the energy consumption behavior of the i-th user in the k-th category of the source domain and the uncertain factors is:

[0033]

[0034] Where, is the adjustable load baseline load of the i-th user in the source domain, are the quadratic regression parameters of the i-th user in the source domain;

[0035] The set of quadratic regression parameters of the energy consumption behavior of the i-th user in the k-th category of the target domain and the uncertain factors is:

[0036]

[0037] Where, is the adjustable load baseline load of the i-th user in the target domain, is the quadratic regression parameter of the i-th user in the target domain;

[0038] The cosine similarity between the quadratic regression parameter of the i-th user in the k-th category of the target domain and the quadratic regression parameter of the j-th user in the k-th category of the source domain is calculated according to the following formula:

[0039]

[0040] Where, are the quadratic regression parameters of the i-th user in the source domain and the j-th user in the target domain, respectively.

[0041] Furthermore, obtaining the probability distribution of the adjustable load adjustment capability of the target domain according to the cosine similarity includes:

[0042] According to the cosine similarity, the target domain user's adjustable load adjustment capability pseudo label is calculated:

[0043]

[0044] Where N is the number of users, cosθ k,i,j is the cosine similarity, is the actual regulation rate of the adjustable load of the source domain user, It is a pseudo label for the load adjustment capability of the target domain user;

[0045] The target domain user adjustable load adjustment capability pseudo label is converted into its probability distribution:

[0046]

[0047] Where, is the probability distribution of the target domain's adjustable load regulation capability.

[0048] Furthermore, aligning the energy consumption behavior characteristics of users in the source domain and the target domain according to the local maximum mean difference based on the probability distribution of the adjustable load regulation capability of the target domain includes:

[0049] The unbiased estimate of the local maximum mean difference obtained by the regenerated Hilbert space kernel function is:

[0050]

[0051]

[0052] Where p and q are the response data set distributions, K is the number of clusters, and D s is the source domain containing the actual adjusted label samples, D t is the target domain containing the actual adjusted label samples, are source domain sample i and target domain sample j respectively, are the weights of the i-th user sample in the source domain belonging to class k and the weights of the j-th user sample in the target domain belonging to class k, respectively. l is the l-th layer of the fully connected layer. is the probability distribution of the load adjustment capability of the target domain users, Regenerate Hilbert space mapping functions for source domain users and target domain users respectively;

[0053] Align the features of the fully connected layers of the source and target domain network models to obtain an unbiased estimate:

[0054]

[0055] Where p and q are the response data set distributions, K is the number of clusters, and N is the number of users. are the weight of the j-th user sample in the source domain belonging to class k and the weight of the i-th user sample in the target domain belonging to class k, respectively. is the lth layer feature of the fully connected layer in the source domain, is the lth layer feature of the fully connected layer of the target domain, and is the kernel function.

[0056] Furthermore, the online transfer learning network model loss function is as follows:

[0057]

[0058] Where N is the number of users, H is the cross entropy loss, and x i 、R i 、c i are the energy consumption behavior, uncertainty factors, and electricity price of the i-th user, respectively; f is the probability prediction label of the adjustable load adjustment amount in the target domain; is the actual adjustment value label of the adjustable load of the i-th user in the source domain, λ is the adaptive loss coefficient, L is the number of fully connected layers, d l '(p,q) is the adaptive loss.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] One of the beneficial effects of the present invention is that, based on the current electricity market environment and diversified interest demands, taking into account the various uncertain factors affecting users' energy consumption behavior, the target domain user's adjustable load regulation capability pseudo-label is obtained according to the cosine similarity of parameter characteristics between users within the class, and the correlation characteristics between typical users' energy consumption behavior and historical regulation data are effectively extracted.

[0061] One of the beneficial effects of the present invention is that, considering different sample weights, the local maximum mean difference is used to measure the empirical probability distribution of relevant subdomains of the source domain and the target domain, and the regenerated Hilbert space kernel function is embedded. On the basis of user classification, the local maximum mean difference is used as the feature alignment indicator for online transfer learning, thereby reducing the negative transfer effect problem in online transfer learning.

[0062] One of the beneficial effects of the present invention is that the energy consumption behavior characteristics of users in the source domain and the target domain are aligned according to the local maximum mean difference, and online transfer learning based on feature alignment is used to realize the evaluation of the adjustable load regulation capability of the target domain, promote the consumption of new energy and the dynamic balance of source and load, ensure the safety and stability of the system, provide theoretical reference and technical support for demand-side management, effectively reduce operational risks, and improve system energy efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 This is a flow chart of a method for evaluating adjustable load regulation capability oriented to power grid regulation needs according to an embodiment of the present invention;

[0064] Figure 2 This is a flowchart of an online transfer learning network model training process according to an embodiment of the present invention. DETAILED DESCRIPTION

[0065] The present invention will be further described below in conjunction with specific examples. The following examples are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention.

[0066] As mentioned above, it is urgent and necessary to incorporate demand-side adjustable load resources into the normalized grid operation and regulation, promote the transformation of the grid from "source follows load" to "source and load interact", and realize the evaluation of the adjustable load regulation capacity.

[0067] To this end, the embodiment of the present invention provides a method for evaluating the adjustable load regulation capability oriented to the power grid regulation demand, such as Figure 1 As shown, the method includes:

[0068] Step 11: Obtain energy consumption data of users to be evaluated on the demand side, wherein the user energy consumption data includes user energy consumption behavior, uncertain factors affecting user energy consumption behavior, and electricity prices;

[0069] In step 12, the energy consumption data of the user to be evaluated is input into the trained online transfer learning network model, and the model outputs the user adjustable load adjustment ratio, that is, the percentage of the adjustment amount to the total load.

[0070] Among them, the online transfer learning network model includes source domain and target domain network models. The source domain and target domain network structures are the same, and are both composed of an input layer, M convolutional layers, L fully connected layers, and an output layer connected in sequence.

[0071] like Figure 2 As shown in Figure 2, the online transfer learning network model is trained by the following method:

[0072] Step 21: Obtain historical energy usage data of the source domain and the target domain;

[0073] Obtain the characteristics of demand-side load resource data, perform vertical stratification based on the grid structure, and combine heuristic dynamic partitioning methods for horizontal partitioning and resource aggregation to divide the demand side into multiple hierarchical partitions. Select one of the hierarchical partitions as the target domain and obtain historical energy consumption data for the target domain.

[0074] In this example, the target domain samples contain a small amount of historical energy usage data for smart buildings, industrial and commercial users, and residents, while the source domain samples used contain a large amount of historical energy usage data for buildings. The target and source domain energy usage data each include user energy usage behavior, various uncertainties that influence user energy usage behavior, and electricity prices.

[0075] Step 22: constructing a source domain adjustable load generalized model and a target domain adjustable load generalized model based on user energy consumption behaviors and uncertain factors affecting user energy consumption behaviors in the source domain and target domain data, respectively; fitting the source domain adjustable load generalized model and the target domain adjustable load generalized model to obtain quadratic regression dynamic relationships between user energy consumption behaviors in the source domain and target domain and multiple uncertain factors, respectively;

[0076] The generalized model of adjustable load is used to describe the relationship between user energy consumption behavior and changes in uncertain factors. In order to reflect the energy consumption curve of the generalized adjustable load that changes over time, its energy consumption curve can be expressed as follows:

[0077] P t =P b,t (R m )+ΔP t (ΔR)=P b,t (R m )(1+f t (ΔR)

[0078] Where, P t is the adjustable load at time t; P b,t (R m) is the adjustable load baseline load at time t, which is related to the uncertainty factor R that affects the user's energy consumption behavior. m Related; ΔP t (ΔR) is the adjustable load change value at time t, which is related to the uncertain factor ΔR that affects the user's energy consumption behavior; R m and ΔR are the average value and variation value of the uncertainty factors in a certain period of time (such as meteorological data, etc.); f t (ΔR) is P b,t (R m ) and ΔP t (ΔR) is a function of the uncertainty factor.

[0079] For users who are sensitive to uncertainties, the quadratic regression equation can be used to characterize the relationship between the change in adjustable load power and uncertainties. The quadratic regression dynamic relationship between energy consumption behavior and various uncertainties can be fitted as follows:

[0080] f t (ΔR)=f 1,t +f 2,t ΔR+f 3,t ΔR 2

[0081] Where, f 1,t 、f 2,t 、f 3,t are the parameters of the quadratic regression.

[0082] Step 23: Cluster the historical energy usage data of the source and target domains to extract similar energy usage behaviors among different users. Based on the clustering results of the source and target domains and the quadratic regression dynamic relationship between the energy usage behaviors of the source and target domain users and various uncertain factors, the quadratic regression parameters of the energy usage behaviors of the source and target domain users within the cluster and the uncertain factors are obtained.

[0083] Among them, the K clustering algorithm is used to extract similar energy consumption behaviors among different users from the historical energy consumption data of the source domain and the target domain.

[0084] Considering the adjustable load characteristics, K sample points are selected, each sample point represents the initial cluster center of each cluster, and then the distance from the remaining samples to the cluster center is calculated and assigned to the nearest cluster. Then the average value of each cluster is recalculated. The whole process is repeated to adjust the cluster center. When the cluster center no longer changes, the clusters formed by data clustering have converged, and K types of user energy consumption behaviors are output. Specifically:

[0085] (1) Randomly select K users’ energy consumption history data as the initial cluster centers;

[0086] (2) Calculate the distance between the remaining samples and the cluster center and assign the samples to the nearest cluster:

[0087]

[0088] Where K is the number of clusters, v ik Indicates whether the i-th user sample belongs to the k-th category, v ik 1 means it belongs to class k, v ik 0 means it does not belong to class k, d(c m ,x i ) is the sample x i To cluster center c m distance;

[0089] (3) Update cluster centers:

[0090]

[0091] Where N is the number of users;

[0092] (4) Determine the convergence condition so that the following formula is minimized:

[0093]

[0094] Step 24: Calculate the cosine similarity of the quadratic regression parameter characteristics of the target domain and source domain users' energy consumption behaviors and uncertain factors within the class, and obtain the probability distribution of the adjustable load regulation capacity of the target domain based on the cosine similarity;

[0095] The cosine similarity of the quadratic regression parameter characteristics of the target domain and source domain users' energy consumption behavior and uncertainty factors within the class is calculated according to the following steps:

[0096] Assume that the quadratic regression parameter set of the energy consumption behavior of the i-th user in the k-th category of the source domain and the uncertain factors is:

[0097]

[0098] Where, is the adjustable load baseline load of the i-th user in the source domain, are the quadratic regression parameters of the i-th user in the source domain;

[0099] The set of quadratic regression parameters of the energy consumption behavior of the i-th user in the k-th category of the target domain and the uncertain factors is:

[0100]

[0101] Where, is the adjustable load baseline load of the i-th user in the target domain, is the quadratic regression parameter of the i-th user in the target domain;

[0102] The cosine similarity between the quadratic regression parameter of the i-th user in the k-th category of the target domain and the quadratic regression parameter of the j-th user in the k-th category of the source domain is calculated according to the following formula:

[0103]

[0104] Where, are the quadratic regression parameters of the i-th user in the source domain and the j-th user in the target domain, respectively.

[0105] The probability distribution of the adjustable load regulation capability of the target domain is obtained according to the cosine similarity, which specifically includes the following steps:

[0106] According to the cosine similarity, the target domain user's adjustable load adjustment capability pseudo label is calculated:

[0107]

[0108] Where N is the number of users, cosθ k,i,j is the cosine similarity, is the actual regulation rate of the adjustable load of the source domain user, It is a pseudo label for the load adjustment capability of the target domain user;

[0109] The target domain user adjustable load adjustment capability pseudo label is converted into its probability distribution:

[0110]

[0111] Where, is the probability distribution of the target domain's adjustable load regulation capability.

[0112] Step 25: Based on the probability distribution of the adjustable load regulation capability of the target domain, the energy consumption behavior characteristics of the users in the source domain and the target domain are aligned according to the local maximum mean difference, and the loss function is minimized by continuously training the model until the model converges.

[0113] This step specifically includes:

[0114] The unbiased estimate of the local maximum mean difference obtained by the regenerated Hilbert space kernel function is:

[0115]

[0116]

[0117] Where p and q are the response data set distributions, K is the number of clusters, and D s is the source domain containing the actual adjusted label samples, D t is the target domain containing the actual adjusted label samples, are source domain sample i and target domain sample j respectively, are the weights of the i-th user sample in the source domain belonging to class k and the weights of the j-th user sample in the target domain belonging to class k, respectively. l is the l-th layer of the fully connected layer. is the probability distribution of the load adjustment capability of the target domain users, Regenerate Hilbert space mapping functions for source domain users and target domain users respectively;

[0118] Align the features of the fully connected layers of the source and target domain network models to obtain an unbiased estimate:

[0119]

[0120] Where p and q are the response data set distributions, K is the number of clusters, and N is the number of users. are the weight of the j-th user sample in the source domain belonging to class k and the weight of the i-th user sample in the target domain belonging to class k, respectively. is the lth layer feature of the fully connected layer in the source domain, is the lth layer feature of the fully connected layer of the target domain, and is the kernel function.

[0121] Among them, the loss function of the online transfer learning network model is as follows:

[0122]

[0123] Where N is the number of users, H is the cross entropy loss, and x i 、R i 、c i are the energy consumption behavior, uncertainty factors, and electricity price of the i-th user, respectively; f is the probability prediction label of the adjustable load adjustment amount in the target domain; is the actual adjustment value label of the adjustable load of the i-th user in the source domain, λ is the adaptive loss coefficient, L is the number of fully connected layers, d′ l (p,q) is the adaptive loss.

[0124] Repeat the above steps and continuously train the model to minimize the loss function until the model converges, obtaining a feature-aligned online transfer learning network model. This model can then be used to evaluate the adjustable load regulation capability of the target domain.

[0125] The present invention has been disclosed above with preferred embodiments, which are not intended to limit the present invention. Any technical solutions obtained by adopting equivalent replacement or equivalent transformation solutions fall within the protection scope of the present invention.

Claims

1. A method for evaluating the adjustable load regulation capability for power grid regulation needs, characterized in that: include: Obtaining energy consumption data of users to be evaluated on the demand side, wherein the user energy consumption data includes user energy consumption behavior, uncertainty factors, and electricity prices; Inputting the energy consumption data of the user to be evaluated into a trained online transfer learning network model, and the model outputting a user-adjustable load adjustment ratio; The online transfer learning network model is trained by the following method: Obtain historical energy consumption data of the source domain and the target domain; A source domain adjustable load generalized model and a target domain adjustable load generalized model are constructed based on user energy consumption behaviors and uncertain factors affecting user energy consumption behaviors in source domain and target domain data, respectively. Based on the source domain adjustable load generalized model and the target domain adjustable load generalized model, quadratic regression dynamic relationships between source domain and target domain user energy consumption behaviors and multiple uncertain factors are obtained by fitting. Clustering the historical energy consumption data of the source and target domains is performed to extract similar energy consumption behaviors among different users. Based on the clustering results of the source and target domains and the quadratic regression dynamic relationship between the energy consumption behaviors of the source and target domain users and various uncertain factors, the quadratic regression parameters of the energy consumption behaviors of the source and target domain users within the class and the uncertain factors are obtained. Calculate the cosine similarity of the quadratic regression parameter characteristics of the target domain and source domain users' energy consumption behavior and uncertain factors within the class, and obtain the probability distribution of the adjustable load regulation capacity of the target domain based on the cosine similarity; Based on the probability distribution of the adjustable load regulation capability of the target domain, the energy consumption behavior characteristics of users in the source domain and the target domain are aligned according to the local maximum mean difference, and the loss function is minimized by continuously training the model until the model converges.

2. The method for evaluating the adjustable load regulation capability according to claim 1, wherein: The generalized model of the adjustable load is: ； Where, P t is the adjustable load at time t; is the adjustable load baseline load at time t, which is related to the uncertainty factor R that affects the user's energy consumption behavior. m related; is the adjustable load change value at time t, which is related to the uncertain factors affecting the user's energy consumption behavior Related; R m and are the average value and change value of the uncertainty factor over a period of time respectively; for and function.

3. The method for evaluating the adjustable load regulation capability according to claim 2, wherein: The quadratic regression dynamic relationship between the user's energy consumption behavior and various uncertain factors is: ； Where, f 1,t 、f 2,t 、f 3,t are the parameters of the quadratic regression.

4. The method for evaluating the adjustable load regulation capability according to claim 1, wherein: The clustering of historical energy usage data in the source domain and the target domain to extract similar energy usage behaviors among different users includes: Randomly select K users’ energy consumption history data as the initial cluster centers; Calculate the distance between the remaining samples and the cluster center and assign the samples to the nearest cluster: ； Where K is the number of clusters, v ik Indicates whether the i-th user sample belongs to the k-th category, v ik 1 means it belongs to class k, v ik 0 means it does not belong to class k. For sample x i To cluster center c m distance; Update cluster centers: ； Where N is the number of users; Determine the convergence condition to minimize the following formula: 。 5. The method for evaluating the adjustable load regulation capability according to claim 3, wherein: The cosine similarity of the quadratic regression parameter characteristics of the target domain and source domain users' energy consumption behaviors and uncertain factors within the class is calculated according to the following steps: Assume that the quadratic regression parameter set of the energy consumption behavior of the i-th user in the k-th category of the source domain and the uncertain factors is: ； Where, is the adjustable load baseline load of the i-th user in the source domain, 、 、 are the quadratic regression parameters of the i-th user in the source domain; The set of quadratic regression parameters of the energy consumption behavior of the i-th user in the k-th category of the target domain and the uncertain factors is: ； Where, is the adjustable load baseline load of the i-th user in the target domain, 、 、 is the quadratic regression parameter of the i-th user in the target domain; The cosine similarity between the quadratic regression parameter of the i-th user in the k-th category of the target domain and the quadratic regression parameter of the j-th user in the k-th category of the source domain is calculated according to the following formula: ； Where, 、 are the quadratic regression parameters of the i-th user in the source domain and the j-th user in the target domain, respectively.

6. The method for evaluating the adjustable load regulation capability according to claim 5, wherein: Obtaining the probability distribution of the adjustable load adjustment capability of the target domain according to the cosine similarity includes: According to the cosine similarity, the target domain user's adjustable load adjustment capability pseudo label is calculated: ； Where N is the number of users, is the cosine similarity, is the actual regulation rate of the adjustable load of the source domain user, It is a pseudo label for the load adjustment capability of the target domain user; The target domain user adjustable load adjustment capability pseudo label is converted into its probability distribution: ； Where, is the probability distribution of the target domain's adjustable load regulation capability.

7. The method for evaluating the adjustable load regulation capability according to claim 6, wherein: The step of aligning the energy consumption behavior characteristics of users in the source domain and the target domain based on the probability distribution of the adjustable load regulation capability of the target domain according to the local maximum mean difference includes: The unbiased estimate of the local maximum mean difference obtained by the regenerated Hilbert space kernel function is: ； ； Where p and q are the response data set distributions, K is the number of clusters, and D s is the source domain containing the actual adjusted label samples, D t is the target domain containing the actual adjusted label samples, 、 are source domain sample i and target domain sample j respectively, 、 are the weights of the i-th user sample in the source domain belonging to class k and the weights of the j-th user sample in the target domain belonging to class k, respectively. l is the l-th layer of the fully connected layer. is the probability distribution of the load adjustment capability of the target domain users, 、 Regenerate Hilbert space mapping functions for source domain users and target domain users respectively; Align the features of the fully connected layers of the source and target domain network models to obtain an unbiased estimate: ； Where p and q are the response data set distributions, K is the number of clusters, and N is the number of users. 、 are the weight of the j-th user sample in the source domain belonging to class k and the weight of the i-th user sample in the target domain belonging to class k, respectively. 、 is the lth layer feature of the fully connected layer in the source domain, 、 is the lth layer feature of the fully connected layer of the target domain, 、 and is the kernel function.

8. The method for evaluating the adjustable load regulation capability according to claim 7, wherein: The online transfer learning network model loss function is as follows: ； Where N is the number of users, H is the cross entropy loss, and x i 、R i 、c i are the energy consumption behavior, uncertainty factors, and electricity price of the i-th user, respectively; f is the probability prediction label of the adjustable load adjustment amount in the target domain; is the actual adjustment value label of the adjustable load of the i-th user in the source domain, is the adaptive loss coefficient, L is the number of fully connected layers, is the adaptive loss.

Images